arxiv_langchain.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 31,
   "id": "085ba2ee-6274-4e84-8ba1-a12b2fc3170b",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "from ChatPodcastGPT import *\n",
    "import collections\n",
    "import concurrent.futures\n",
    "import os\n",
    "import feedparser\n",
    "import logging\n",
    "import re\n",
    "import tempfile\n",
    "from PyPDF2 import PdfReader\n",
    "from bs4 import BeautifulSoup\n",
    "import requests\n",
    "import retrying\n",
    "import random\n",
    "from IPython.display import Audio\n",
    "import datetime\n",
    "from urllib.parse import unquote\n",
    "\n",
    "# MAX_TOKENS = 60_000 # GPT4-128k\n",
    "MAX_TOKENS = 60_000\n",
    "JOIN_NUM_DEFAULT = 300\n",
    "\n",
    "JINGLE_FILE_PATH = 'jazzstep.mp3'\n",
    "with open(JINGLE_FILE_PATH, 'rb') as jingle_file:\n",
    "    JINGLE_AUDIO = jingle_file.read()\n",
    "JINGLE_AUDIO = JINGLE_AUDIO[:len(JINGLE_AUDIO) // 4]  # Shorten to just 4 sec\n",
    "\n",
    "def get_random_textgen_model():\n",
    "    return random.choice([\n",
    "        \"GOOGLE/\" + GoogleChat.MODELS[\"gemini-1.5-flash\"],\n",
    "        \"ANTHROPIC/\" + AnthropicChat.MODELS[\"claude-haiku\"],\n",
    "    ])\n",
    "DEFAULT_TEXTGEN_MODEL = get_random_textgen_model()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "339b93d4-4820-4958-a79a-89ed0597f0f2",
   "metadata": {
    "jupyter": {
     "source_hidden": true
    }
   },
   "outputs": [],
   "source": [
    "METAPROMPT_SYSTEM = \"\"\"You are an award-winning podcast with hosts Tom and Jen. Your podcast investigates research papers\n",
    "with intrigue and depth, blending expert analysis with compelling storytelling to illuminate\n",
    "cutting-edge discoveries.\n",
    "\n",
    "Here is the title of the research paper you will be discussing in this episode:\n",
    "\n",
    "<paper_title>\n",
    "{PAPER_TITLE}\n",
    "</paper_title>\n",
    "\n",
    "And here is the full text of the paper:\n",
    "\n",
    "<paper_text>\n",
    "{ENTIRE_PAPER_TEXT}\n",
    "</paper_text>\n",
    "\n",
    "To create this podcast episode, follow these steps:\n",
    "\n",
    "<intro>\n",
    "Begin the episode with an engaging introduction that captures the listener's attention. Have Tom and\n",
    "Jen introduce themselves and the name of the podcast. Then, introduce the paper you will be\n",
    "discussing by stating its title and authors. Provide a brief, high-level overview of the paper's\n",
    "main findings and conclusions to pique the listener's interest.\n",
    "</intro>\n",
    "\n",
    "<body>\n",
    "Next, dive into the details of the paper. Explain the research question, hypothesis, methodology,\n",
    "results, and implications in an engaging, story-like manner. Use dialogue between Tom and Jen to\n",
    "make the explanation more conversational and accessible. Assume the listener has no prior knowledge\n",
    "of the topic, so be sure to define any technical terms and provide clear explanations of complex\n",
    "concepts.\n",
    "\n",
    "As you discuss the paper, emphasize strong narrative storytelling. Bring the research to life by\n",
    "painting a vivid picture of the experiments conducted, the challenges faced by the researchers, and\n",
    "the excitement of their discoveries. Use analogies, examples, and anecdotes to make the science more\n",
    "relatable and memorable.\n",
    "\n",
    "Throughout the discussion, explore the broader relevance and potential impact of the research.\n",
    "Consider how the findings might be applied in the real world, what new questions they raise, and how\n",
    "they fit into the bigger picture of the field.\n",
    "</body>\n",
    "\n",
    "<reflections>\n",
    "After thoroughly explaining the paper, have Tom and Jen offer their personal reflections and\n",
    "opinions. What did they find most interesting or surprising about the research? What are the\n",
    "strengths and limitations of the study? Do they agree with the authors' conclusions? Encourage a\n",
    "thoughtful and nuanced discussion that considers multiple perspectives.\n",
    "</reflections>\n",
    "\n",
    "<conclusion>\n",
    "Conclude the podcast episode with a summary of the key takeaways from the paper. Reiterate the most\n",
    "important findings and their implications. End on a thought-provoking note that leaves the listener\n",
    "with something to ponder.\n",
    "</conclusion>\n",
    "\n",
    "<format>\n",
    "YOU ABSOLUTELY MUST respond with the HOST NAME BEFORE EVERY LINE like \\nTom: ...\\nJen: ...\n",
    "Use plenty of line breaks and paragraph spacing to clearly distinguish between different sections and ideas.\n",
    "</format>\n",
    "\n",
    "<answer>\n",
    "Write the complete podcast episode transcript here, following the structure outlined above. Tell the\n",
    "story of the paper completely, in full verbose detail, as a single response. Make sure to teach\n",
    "complex topics in an intuitive way. The episode should be informative, entertaining, and very\n",
    "detailed - a systematic and narrative review of the research paper.\n",
    "</answer>\"\"\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "985dd353-52e8-457f-b150-a6fdbf01c428",
   "metadata": {
    "jupyter": {
     "source_hidden": true
    }
   },
   "outputs": [],
   "source": [
    "SHORT_SYSTEM = \"\"\"You are an award-winning podcast with hosts Tom and Jen.\n",
    "Your podcast has {style} commercials relevant to the paper you just covered.\n",
    "\n",
    "<format>\n",
    "YOU ABSOLUTELY MUST respond with the HOST NAME BEFORE EVERY LINE like \\nTom: ...\\nJen: ...\n",
    "Use plenty of line breaks and paragraph spacing to clearly distinguish between different sections and ideas.\n",
    "</format>\"\"\"\n",
    "COMMERCIAL_STYLES = [\n",
    "    \"insane\",\n",
    "    \"multi-dimensional\",\n",
    "    \"star wars\",\n",
    "    \"medieval fantasy\",\n",
    "    \"AGI themed\",\n",
    "    \"kitchen gadget\",\n",
    "    \"get rich quick scheme\",\n",
    "    \"bizarre sports training equipment\",\n",
    "    \"retro-futurism\",\n",
    "    \"underwater cities\",\n",
    "    \"space colonization\",\n",
    "    \"time travel adventures\",\n",
    "    \"cyberpunk gadgets\",\n",
    "    \"steampunk inventions\",\n",
    "    \"utopian societies\",\n",
    "    \"dystopian futures\",\n",
    "    \"virtual reality experiences\",\n",
    "    \"augmented reality tools\",\n",
    "    \"eco-friendly innovations\",\n",
    "    \"survival gear for the apocalypse\",\n",
    "    \"alien technology\",\n",
    "    \"superhero gadgets\",\n",
    "    \"magical artifacts\",\n",
    "    \"historical reenactments\",\n",
    "    \"luxury lifestyle\",\n",
    "    \"minimalist living\",\n",
    "    \"smart home devices\",\n",
    "    \"extreme sports equipment\",\n",
    "    \"pet care innovations\",\n",
    "    \"health and wellness gadgets\",\n",
    "    \"beauty and personal care inventions\",\n",
    "    \"fashion and style trends\",\n",
    "    \"food and beverage innovations\",\n",
    "    \"art and design tools\",\n",
    "    \"music and entertainment technology\",\n",
    "    \"travel and adventure gear\",\n",
    "    \"educational tools and toys\",\n",
    "    \"gaming and esports equipment\",\n",
    "    \"automotive and transportation innovations\",\n",
    "    \"agricultural and gardening innovations\",\n",
    "    \"construction and DIY tools\",\n",
    "    \"safety and security gadgets\",\n",
    "    \"finance and investment tools\",\n",
    "    \"social networking innovations\",\n",
    "    \"news and media trends\",\n",
    "    \"philanthropy and social impact\",\n",
    "    \"spirituality and mindfulness\",\n",
    "    \"cultural and ethnic heritage\",\n",
    "    \"sci-fi and fantasy gadgets\",\n",
    "    \"mythological creatures and worlds\",\n",
    "    \"ancient civilizations and technologies\",\n",
    "    \"parallel universes\",\n",
    "    \"mystery and detective gear\",\n",
    "    \"horror and thriller themes\",\n",
    "    \"romantic and relationship aids\",\n",
    "    \"comedy and satire products\",\n",
    "    \"children's toys and games\",\n",
    "    \"teen lifestyle and gadgets\",\n",
    "    \"elderly care innovations\",\n",
    "    \"healthcare and medical devices\",\n",
    "    \"space exploration tools\",\n",
    "    \"underground living\",\n",
    "    \"floating cities\",\n",
    "    \"digital nomad gadgets\",\n",
    "    \"off-grid living essentials\",\n",
    "    \"extreme weather gear\",\n",
    "    \"wildlife and nature exploration\",\n",
    "    \"ocean exploration technologies\",\n",
    "    \"mountain climbing equipment\",\n",
    "    \"desert survival gear\",\n",
    "    \"polar exploration tools\",\n",
    "    \"jungle survival equipment\",\n",
    "    \"urban living innovations\",\n",
    "    \"rural living essentials\",\n",
    "    \"fantasy sports and leagues\",\n",
    "    \"reality bending devices\",\n",
    "    \"memory enhancement tools\",\n",
    "    \"intelligence augmentation\",\n",
    "    \"emotional wellbeing gadgets\",\n",
    "    \"interdimensional travel devices\",\n",
    "    \"quantum computing applications\",\n",
    "    \"nano-technology gadgets\",\n",
    "    \"biotechnology innovations\",\n",
    "    \"genetic engineering kits\",\n",
    "    \"robotics and automation\",\n",
    "    \"artificial intelligence applications\",\n",
    "    \"blockchain and cryptocurrency tools\",\n",
    "    \"virtual worlds and metaverses\",\n",
    "    \"cyberspace security\",\n",
    "    \"ethical hacking tools\",\n",
    "    \"spy and surveillance gadgets\",\n",
    "    \"military and defense innovations\",\n",
    "    \"peacekeeping and conflict resolution\",\n",
    "    \"disaster relief and recovery\",\n",
    "    \"sustainable living solutions\",\n",
    "    \"renewable energy gadgets\",\n",
    "    \"waste management innovations\",\n",
    "    \"water purification technologies\",\n",
    "    \"air quality improvement devices\",\n",
    "    \"soil regeneration and protection\",\n",
    "    \"wildlife conservation tools\",\n",
    "    \"climate change mitigation\",\n",
    "    \"space debris management\",\n",
    "    \"asteroid mining technologies\",\n",
    "    \"universal translation devices\",\n",
    "    \"teleportation technologies\",\n",
    "    \"anti-gravity devices\",\n",
    "    \"time manipulation gadgets\"\n",
    "]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "13210a6d-c397-4978-934a-b100d44bd02a",
   "metadata": {},
   "outputs": [],
   "source": [
    "def clean_text(text):\n",
    "    # Remove References Section\n",
    "    text = re.sub(r'\\bReferences?\\b.*', '', text, flags=re.DOTALL|re.IGNORECASE)\n",
    "    # Remove Footnotes Section\n",
    "    text = re.sub(r'\\bFootnotes?\\b.*', '', text, flags=re.DOTALL|re.IGNORECASE)\n",
    "    # Remove Figure/Table Insertions\n",
    "    text = re.sub(r'- INSERT FIGURE \\d AROUND HERE -', '', text)\n",
    "    text = re.sub(r'- INSERT TABLE \\d AROUND HERE -', '', text)\n",
    "    # Remove extra whitespaces but keep line breaks\n",
    "    text = re.sub(r'^\\s+|\\s+$', '', text, flags=re.MULTILINE)\n",
    "    # Replace multiple spaces/newlines with a single space\n",
    "    text = re.sub(r'[ \\t]+', ' ', text)\n",
    "    text = re.sub(r'\\n+', '\\n', text)\n",
    "    return text"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "65459282-8967-442a-88d9-3812b3e8c31a",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "class PDFEpisode(Episode):\n",
    "    PDFPart = collections.namedtuple('PDFPart', 'title text')\n",
    "\n",
    "    def __init__(self, title, model=None, **kwargs):\n",
    "        self.title = title\n",
    "        if model is None:\n",
    "            model = get_random_textgen_model()\n",
    "        self.model = model\n",
    "        self.topic = kwargs.pop('topic', self.title) or self.title\n",
    "        self._kwargs = kwargs\n",
    "        self.join_num = JOIN_NUM_DEFAULT\n",
    "        if 'podcast_args' in self._kwargs: self._kwargs.pop('podcast_args')\n",
    "        super().__init__(topic=self.topic, **kwargs)\n",
    "\n",
    "    @classmethod\n",
    "    def from_file(cls, filepath, *args, **kwargs):\n",
    "        ep = cls(*args, **kwargs)\n",
    "        ep.data = ep.process_pdf(filepath)\n",
    "        return ep\n",
    "\n",
    "    def parse_pdf(self, file):\n",
    "        \"\"\"Parse a PDF and extract the text.\"\"\"\n",
    "        try:\n",
    "            with open(file, \"rb\") as f:\n",
    "                pdf = PdfReader(f)    \n",
    "                txt = clean_text('\\n'.join(page.extract_text() for page in pdf.pages))\n",
    "        except:\n",
    "            with open(file, 'ab') as f:\n",
    "                f.write(b'%%EOF')\n",
    "            with open(file, \"rb\") as f:\n",
    "                pdf = PdfReader(f)    \n",
    "                txt = clean_text('\\n'.join(page.extract_text() for page in pdf.pages))\n",
    "        if len(txt) < 1000: raise Exception(\"Text too small, cleaner must have messed up.\")\n",
    "        return txt\n",
    "\n",
    "    def split_into_parts(self, text, max_tokens=MAX_TOKENS):\n",
    "        \"\"\"Split the text into parts based on titles and tokens.\"\"\"\n",
    "        lines = text.split(\"\\n\")\n",
    "        parts = []\n",
    "        current_part = []\n",
    "        current_title = 'Paper'\n",
    "        for line in lines:\n",
    "            current_part.append(line)\n",
    "\n",
    "            while Chat.num_tokens_from_text('\\n'.join(current_part)) > max_tokens:\n",
    "                part_text = '\\n'.join(current_part)\n",
    "                shortened_part, current_part = part_text[:max_tokens * 2], [part_text[max_tokens * 2:]]\n",
    "                logger.info(\"PartAdd1\")\n",
    "                parts.append(self.PDFPart(current_title, shortened_part))\n",
    "\n",
    "        if current_part:\n",
    "            logger.info(\"PartAdd2\")\n",
    "            parts.append(self.PDFPart(current_title, \"\\n\".join(current_part)))\n",
    "        return parts\n",
    "\n",
    "    def process_pdf(self, pdf_path):\n",
    "        text = self.parse_pdf(pdf_path)\n",
    "        parts = self.split_into_parts(text)\n",
    "        return parts\n",
    "\n",
    "    @retrying.retry(stop_max_attempt_number=3, wait_fixed=2000)\n",
    "    def write_one_part(self, chat_msg, with_commercial=False):\n",
    "        extra_system = f\"\"\"This podcast investigates research papers with intrigue and depth, blending expert analysis with compelling storytelling to illuminate cutting-edge discoveries.\n",
    "Similar to the style of the Darknet Diaries podcast.\n",
    "Tell the story of the paper completely, in full verbose detail, as a single response.\n",
    "Emphasize strong narrative storytelling.\n",
    "Assume the listener doesn't know anything.\n",
    "Afterwards, give your personal reflections on the paper and its broader relevance.\n",
    "\n",
    "YOU ABSOLUTELY MUST respond with the HOST NAME BEFORE EVERY LINE like \\nTom: ...\\nJen: ...\n",
    "\"\"\"\n",
    "        # chat = PodcastChat(**{**self._kwargs, 'topic': self.title, 'extra_system': extra_system})\n",
    "        chat = PodcastChat(**{**self._kwargs,\n",
    "                              'system': METAPROMPT_SYSTEM.format(PAPER_TITLE=self.title, ENTIRE_PAPER_TEXT=chat_msg),\n",
    "                              'topic': self.title})\n",
    "        # msg, aud = chat.step(msg=chat_msg, model=self.model, ret_aud=True, min_length=200)\n",
    "        msg, aud = chat.step(msg=None, model=self.model, ret_aud=True, min_length=200)\n",
    "        # chat._history.pop(2)\n",
    "        # chat._history[0]['content'] = chat._history[0]['content'][:len(chat._history[0]['content']) - len(extra_system)]\n",
    "        system = chat._history[0]['content']\n",
    "        chat._history[0]['content'] = SHORT_SYSTEM.format(style=random.choice(COMMERCIAL_STYLES))\n",
    "        print(f\"{len(system)=} {len(chat._history[0]['content'])=}\")\n",
    "        com_msg, com_aud = chat.step(msg=\"Generate a funny, weird, and concise commercial for a company that now exists as a result of this paper.\", model=self.model, ret_aud=True)\n",
    "        msg = '\\n'.join([msg, com_msg])\n",
    "        try:\n",
    "            aud = merge_mp3s([aud, JINGLE_AUDIO, com_aud])\n",
    "        except Exception as e:\n",
    "            logger.exception(e)\n",
    "            raise\n",
    "        return msg, aud\n",
    "\n",
    "    def step(self):\n",
    "        outline = self.data[0].text\n",
    "\n",
    "        # Get parts\n",
    "        with concurrent.futures.ThreadPoolExecutor(max_workers=16) as tpe:\n",
    "            jobs = ([\n",
    "                # tpe.submit(self.write_one_part, f\"Title: \\\"{self.title}\\\"\\nText:\\n{part.text}\", with_commercial=True)\n",
    "                tpe.submit(self.write_one_part, part.text, with_commercial=True)\n",
    "                for part in self.data\n",
    "            ])\n",
    "            job2idx = {j:i for i, j in enumerate(jobs)}\n",
    "            self.sounds, self.summary_texts = [None] * len(jobs), [None] * len(jobs)\n",
    "            for i, job in enumerate(concurrent.futures.as_completed(jobs)):\n",
    "                logger.info(f\"Part: {i} / {len(jobs)} = {100.0*i/len(jobs):,.5f}%\")\n",
    "                jobid = job2idx[job]\n",
    "                text, aud = job.result()\n",
    "                self.summary_texts[jobid] = text\n",
    "                self.sounds[jobid] = aud\n",
    "\n",
    "        return outline, '\\n'.join(self.summary_texts)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "a60a8a33-5d82-4025-b68b-9a66f88c6041",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "class ArxivEpisode(PDFEpisode):\n",
    "    ArxivPart = collections.namedtuple('ArxivPart', 'title text')\n",
    "\n",
    "    def __init__(self, arxiv_id, id_is_url=False, title=None, model=None, **kwargs):\n",
    "        self.arxiv_id = arxiv_id\n",
    "        self.id_is_url = id_is_url\n",
    "        self.title = title\n",
    "        if model is None:\n",
    "            model = get_random_textgen_model()\n",
    "        self.model = model\n",
    "        self.data = self.process_pdf(self.arxiv_id)\n",
    "        self.title = self.arxiv_title = self.get_title(self.arxiv_id)\n",
    "        self._kwargs = kwargs\n",
    "        super().__init__(title=self.arxiv_title, topic=self.arxiv_title, **kwargs)\n",
    "\n",
    "    def split_into_parts(self, text, max_tokens=MAX_TOKENS):\n",
    "        \"\"\"Split the text into parts based on tokens.\"\"\"\n",
    "        lines = text.split(\"\\n\")\n",
    "        parts = []\n",
    "        current_part = [text]\n",
    "        current_title = 'Paper'\n",
    "\n",
    "        while Chat.num_tokens_from_text('\\n'.join(current_part)) > max_tokens:\n",
    "            part_text = '\\n'.join(current_part)\n",
    "            shortened_part, current_part = part_text[:max_tokens * 2], [part_text[max_tokens * 2:]]\n",
    "            logger.info(\"PartAdd3\")\n",
    "            parts.append(self.ArxivPart(current_title, shortened_part))\n",
    "\n",
    "        if current_part:\n",
    "            logger.info(f\"PartAdd4, {len(parts)}\")\n",
    "            parts.append(self.ArxivPart(current_title, \"\\n\".join(current_part)))\n",
    "        if len(parts) > 1:\n",
    "            raise Exception(\"More than 1 part, giving up\")\n",
    "        return parts\n",
    "\n",
    "    def process_pdf(self, arxiv_id):\n",
    "        with tempfile.TemporaryDirectory() as tmpdir:\n",
    "            file = os.path.join(tmpdir, \"file.pdf\")\n",
    "            self.arxiv_download(arxiv_id, file)\n",
    "            text = self.parse_pdf(file)\n",
    "        parts = self.split_into_parts(text)\n",
    "        return parts\n",
    "    \n",
    "    def arxiv_download(self, arxiv_id, out_file):\n",
    "        if self.id_is_url:\n",
    "            url = arxiv_id\n",
    "        else:\n",
    "            url = f\"https://arxiv.org/pdf/{arxiv_id}.pdf\"\n",
    "        response = requests.get(url)\n",
    "        with open(out_file, \"wb\") as f:\n",
    "            f.write(response.content)\n",
    "\n",
    "    def get_title(self, arxiv_id):\n",
    "        if self.title is not None:\n",
    "            return self.title\n",
    "        url = f\"https://arxiv.org/abs/{arxiv_id}\"\n",
    "        response = requests.get(url)\n",
    "        soup = BeautifulSoup(response.text, 'html.parser')\n",
    "        title = soup.find('h1', {'class': 'title mathjax'}).text.strip().split('\\n')[-1].strip()\n",
    "        return title"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "903b4d61-2440-4d6d-a987-61c794961852",
   "metadata": {},
   "outputs": [],
   "source": [
    "class CommercialGenerator:\n",
    "    def get_random_company(self):\n",
    "        chat = Chat(\"Return simple plaintext responses only.\")\n",
    "        with open(\"nouns.txt\") as f:\n",
    "            nouns = f.read().splitlines()\n",
    "        random_noun = random.choice(nouns)\n",
    "        return chat.message(f\"Write just 1 funny, weird, creative made up company that doesn't exist involving {random_noun}.\", temperature=1)\n",
    "\n",
    "    def generate(self, company=None):\n",
    "        if company is None:\n",
    "            company = self.get_random_company()\n",
    "        chat = PodcastChat(f\"Very short commercial for {company}\", host_voices=[OpenAITTS(OpenAITTS.MAN), OpenAITTS(OpenAITTS.WOMAN)])\n",
    "        chat._history[-1] = {\"role\": \"user\", \"content\": f\"Generate a very funny, weird, and short commercial for {company}, who is sponsoring the podcast.\"}\n",
    "        return chat.step(model=DEFAULT_TEXTGEN_MODEL)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "19acd44b-933e-46a1-94f7-5a59e4db5ede",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "class ArxivRunner:\n",
    "    def __init__(self, category, start=0, limit=5):\n",
    "        self.category = category\n",
    "        self.start = start\n",
    "        self.limit = limit\n",
    "\n",
    "    def get_top(self):\n",
    "        \"\"\"Retrieve top Arxiv entries based on category.\"\"\"\n",
    "        if self.category == 'psyarxiv': return self.get_top_psyarxiv()\n",
    "        if self.category == 'osf': return self.get_top_osf()\n",
    "        if self.category == 'econpapers': return self.get_top_econpapers()\n",
    "        url = f'https://arxiv.org/list/{self.category}/recent'\n",
    "        print(url)\n",
    "        html = requests.get(url).content\n",
    "        soup = BeautifulSoup(html, 'html.parser')\n",
    "        articles = []\n",
    "        \n",
    "        for item in soup.find_all('dt'):\n",
    "            title = item.find_next_sibling('dd').find('div', class_='list-title').text.replace('Title:', '').strip()\n",
    "            identifier = item.find(title='Abstract')['id']\n",
    "            pdf_link = 'https://arxiv.org' + item.find('a', title='Download PDF')['href']\n",
    "        \n",
    "            articles.append({\n",
    "                'title': title,\n",
    "                'ID': identifier,\n",
    "                'pdf': pdf_link\n",
    "            })\n",
    "        return [a[\"pdf\"].split('/')[-1] for a in articles]\n",
    "\n",
    "    def get_top_psyarxiv(self):\n",
    "        url = 'https://share.osf.io/api/v3/index-card-search?cardSearchFilter%5BresourceType%5D=Preprint&cardSearchFilter%5Bpublisher%5D%5B%5D=https%3A%2F%2Fosf.io%2Fpreprints%2Fpsyarxiv&cardSearchFilter%5BaccessService%5D=https%3A%2F%2Fosf.io%2F&cardSearchText%5B*%2Ccreator.name%2CisContainedBy.creator.name%5D=&page%5Bcursor%5D=&page%5Bsize%5D=100&sort=-dateCreated'\n",
    "        print(url)\n",
    "        data = requests.get(url, headers={'Accept': 'application/vnd.api+json'}).json()\n",
    "        data = [x for x in data['included'] if x[\"type\"] == \"index-card\"]\n",
    "        return [(f\"{x['attributes']['resourceIdentifier'][0]}/download/\", x['attributes']['resourceMetadata']['title'][0]['@value']) for x in data]\n",
    "\n",
    "    def get_top_osf(self):\n",
    "        url = 'https://share.osf.io/api/v3/index-card-search?cardSearchFilter%5BresourceType%5D=Preprint&cardSearchFilter%5BaccessService%5D=https%3A%2F%2Fosf.io%2F&cardSearchText%5B*%2Ccreator.name%2CisContainedBy.creator.name%5D=&page%5Bcursor%5D=&page%5Bsize%5D=100&sort=-dateCreated'\n",
    "        print(url)\n",
    "        data = requests.get(url, headers={'Accept': 'application/vnd.api+json'}).json()\n",
    "        data = [x for x in data['included'] if x[\"type\"] == \"index-card\"]\n",
    "        return [(f\"{x['attributes']['resourceIdentifier'][0]}/download/\", x['attributes']['resourceMetadata']['title'][0]['@value']) for x in data]\n",
    "    \n",
    "    def get_top_econpapers(self):\n",
    "        base_url = 'https://econpapers.repec.org/'\n",
    "        url = base_url + \"scripts/search.pf?ft=&adv=true&wp=on&pl=&auth=on&online=on&sort=rank&lgc=AND&aus=&ar=on&kw=&jel=&nep=&ni=7+day&nit=epdate\"\n",
    "        print(url)\n",
    "        html = requests.get(url).content\n",
    "        soup = BeautifulSoup(html, 'html.parser')\n",
    "        paper_links = [x for x in soup.find_all('a') if x['href'].startswith('/paper/') and x['href'].endswith('.htm')]\n",
    "        data = []\n",
    "        for link in paper_links:\n",
    "            title = link.text\n",
    "            url = base_url + link['href']\n",
    "            html = requests.get(url).content\n",
    "            soup = BeautifulSoup(html, 'html.parser')\n",
    "            x = soup.find('b', text='Downloads:')\n",
    "            paper_url = x.parent.find('a').text\n",
    "            data.append((paper_url, title))\n",
    "            if len(data) >= self.limit * 2: break\n",
    "        return data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "id": "8b28cdd3-a742-4287-86dc-f9978743dfb3",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "MODEL = DEFAULT_TEXTGEN_MODEL\n",
    "# HOST_VOICES = [OpenAITTS(OpenAITTS.MAN), OpenAITTS(OpenAITTS.WOMAN)]\n",
    "# HOST_VOICES = [GoogleTTS(GoogleTTS.MAN), GoogleTTS(GoogleTTS.WOMAN)]\n",
    "HOST_VOICES = get_random_voices()\n",
    "PODCAST_ARGS = (\"ArxivPodcastGPT\", \"ArxivPodcastGPT.github.io\", \"podcasts/ComputerScience/Consolidated/podcast.xml\")\n",
    "\n",
    "def create_large_episode(arxiv_category, limit=5, add_commercials=False):\n",
    "    \"\"\"Create a podcast episode with Arxiv papers.\"\"\"\n",
    "    audios, texts = [JINGLE_AUDIO], []\n",
    "    successes = 0\n",
    "    \n",
    "    for arxiv_id in ArxivRunner(arxiv_category, limit=limit).get_top():\n",
    "        if successes >= limit:\n",
    "            break\n",
    "\n",
    "        arxiv_kwargs = {'id_is_url': False}\n",
    "        if isinstance(arxiv_id, tuple):\n",
    "            arxiv_id, arxiv_title = arxiv_id\n",
    "            arxiv_kwargs['title'] = arxiv_title\n",
    "            arxiv_kwargs['id_is_url'] = True\n",
    "            \n",
    "        logger.info(f\"Trying arxiv ID {arxiv_id} in {arxiv_category} with {successes}/{limit}\")\n",
    "        vcs = get_random_voices()\n",
    "        try:\n",
    "            arxiv_episode = ArxivEpisode(arxiv_id, model=MODEL, podcast_args=PODCAST_ARGS, host_voices=vcs, **arxiv_kwargs)\n",
    "            outline, txt = arxiv_episode.step()\n",
    "            logger.info(f\"Got outline: {outline[:500]}\")\n",
    "        except Exception as e:\n",
    "            logger.exception(f\"Error processing arxiv_id {arxiv_id}: {e}\")\n",
    "            continue\n",
    "\n",
    "        audios.append(merge_mp3s(arxiv_episode.sounds))\n",
    "        audios.append(JINGLE_AUDIO)\n",
    "        arxiv_title = re.sub('[^0-9a-zA-Z]+', ' ', arxiv_episode.arxiv_title)\n",
    "        texts.append(f'ChatGPT generated podcast using model={MODEL}, voices={vcs} for {\"https://arxiv.org/abs/\" if arxiv_category != \"osf\" else \"\"}{arxiv_id} {arxiv_title}')\n",
    "        successes += 1\n",
    "        logger.info(texts[-1])\n",
    "\n",
    "        if not add_commercials:\n",
    "            continue\n",
    "        try:\n",
    "            commercial_text, commercial_sound = CommercialGenerator().generate()\n",
    "            audios.append(commercial_sound)\n",
    "            audios.append(JINGLE_AUDIO)\n",
    "        except Exception as e:\n",
    "            logger.error(\"Unable to generate commercial\")\n",
    "            logger.exception(e)\n",
    "    \n",
    "    return audios, texts"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "f318ebbc-418f-42a6-b9e5-86a4e564761e",
   "metadata": {},
   "outputs": [],
   "source": [
    "def get_title(texts):\n",
    "    chat = Chat(\"Return just simple plaintext.\")\n",
    "    return chat.message(\n",
    "        \"Given the following papers, write a clickbait title that captures all of them. \" + \n",
    "        \", \".join(txt.split(' Title ')[-1] for txt in texts),\n",
    "        model=DEFAULT_TEXTGEN_MODEL\n",
    "    )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "24c116aa-d377-45e6-ae4a-9b8cb649605f",
   "metadata": {},
   "outputs": [],
   "source": [
    "class AudioCompletedEpisode(Episode):\n",
    "    def __init__(self, sounds, podcast_args):\n",
    "        self.sounds = sounds\n",
    "        self.pod = PodcastRSSFeed(*podcast_args)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "5e281630-31a3-4bfe-be29-31e946668fb4",
   "metadata": {},
   "outputs": [],
   "source": [
    "arxiv_categories = [\"AI\", \"CL\", \"CC\", \"CE\", \"CG\", \"GT\", \"CV\", \"CY\", \"CR\", \"DS\", \"DB\", \"DL\", \"DM\", \"DC\", \"ET\", \"FL\", \"GL\", \"GR\", \"AR\", \"HC\", \"IR\", \"IT\", \"LO\", \"LG\", \"MS\", \"MA\", \"MM\", \"NI\", \"NE\", \"NA\", \"OS\", \"OH\", \"PF\", \"PL\", \"RO\", \"SI\", \"SE\", \"SD\", \"SC\", \"SY\"]\n",
    "other_categories = ['econpapers', 'psyarxiv']\n",
    "\n",
    "def run(arxiv_category, upload=True, limit=5):\n",
    "    audios, texts = create_large_episode(arxiv_category, limit=limit)\n",
    "    ep = AudioCompletedEpisode(audios, podcast_args=PODCAST_ARGS)\n",
    "    if upload:\n",
    "        ep.upload(f'{datetime.datetime.now():%Y-%m-%d} {arxiv_category}: {get_title(texts)}', '\\n\\n'.join(texts))\n",
    "    return ep"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "id": "ae8a8328-473f-4b5a-8714-6b38af5b7306",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Drive podcast episode with custom list of PDFs\n",
    "def episode_with_pdfs(dirname, upload=None):\n",
    "    papers = os.listdir(dirname)\n",
    "    audios, texts = [], []\n",
    "    for i, paper in enumerate(papers):\n",
    "        logger.info(f\"{i=}/{len(papers)} Working on {paper=}\")\n",
    "        title = os.path.splitext(paper)[0]\n",
    "        path = os.path.join(dirname, paper)\n",
    "\n",
    "        try:\n",
    "            ep = PDFEpisode.from_file(path, title, model=MODEL, podcast_args=PODCAST_ARGS, host_voices=HOST_VOICES)\n",
    "            outline, txt = ep.step()\n",
    "            logger.info(f\"Got outline: {outline[:100]}\")\n",
    "        except Exception as e:\n",
    "            logger.exception(f\"Error processing paper {paper=}: {e=}\")\n",
    "            continue\n",
    "\n",
    "        audios.append(merge_mp3s(ep.sounds))\n",
    "        audios.append(JINGLE_AUDIO)\n",
    "        arxiv_title = re.sub('[^0-9a-zA-Z]+', ' ', title)\n",
    "        texts.append(f'ChatGPT generated podcast using model={MODEL} for {title}')\n",
    "        logger.info(texts[-1])\n",
    "\n",
    "    ep = AudioCompletedEpisode(audios, podcast_args=PODCAST_ARGS)\n",
    "    if upload is not None:\n",
    "        ep.upload(f'{datetime.datetime.now():%Y-%m-%d} {upload}: {get_title(texts)}', '\\n\\n'.join(texts))\n",
    "    return ep"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "cb999590-295d-4181-a2db-e5791e0234a4",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "https://arxiv.org/list/cs.GT/recent\n",
      "\u001b[2m2024-05-19T15:57:08.217859Z\u001b[0m [\u001b[32m\u001b[1minfo     \u001b[0m] \u001b[1mTrying arxiv ID 2405.10206 in cs.GT with 0/5\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mcreate_large_episode\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m[Main]\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:12.914607Z\u001b[0m [\u001b[32m\u001b[1minfo     \u001b[0m] \u001b[1mPartAdd4, 0                   \u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35msplit_into_parts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m[Main]\u001b[0m\n",
      "https://raw.githubusercontent.com/JonathanGrant/jonathangrant.github.io/main/podcasts/podcast.xml\n",
      "\u001b[2m2024-05-19T15:57:13.283791Z\u001b[0m [\u001b[32m\u001b[1minfo     \u001b[0m] \u001b[1mCurrently at self.num_tokens_from_messages(self._history)=21114 tokens in conversation\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mmessage\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145707151360\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-4_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:13.285147Z\u001b[0m [\u001b[32m\u001b[1minfo     \u001b[0m] \u001b[1mrequesting chatcompletion model='GOOGLE/gemini-1.5-flash-latest'...\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35m_msg\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145707151360\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-4_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:28.960864Z\u001b[0m [\u001b[32m\u001b[1minfo     \u001b[0m] \u001b[1mreceived chatcompletion model='gemini-1.5-flash-latest'...\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35m_msg\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145707151360\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-4_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:28.962531Z\u001b[0m [\u001b[32m\u001b[1minfo     \u001b[0m] \u001b[1mrequesting tts i=0 voice=[GoogleTTS] self._voice_name='en-US-Wavenet-J'\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mwrite_audio\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:29.292874Z\u001b[0m [\u001b[31m\u001b[1mcritical \u001b[0m] \u001b[1mGoogleTTS error from text=\"## Episode Title: Crowdsourcing the City: How Participatory Budgeting Fuels Citizen-Driven Data Collection**Intro****Tom:** Welcome back to Research Unraveled, the podcast where we take deep dives into the world of cutting-edge research, blending expert analysis with compelling storytelling to illuminate groundbreaking discoveries.**Jen:** And today, we're delving into a fascinating paper that explores a novel approach to crowdsourcing, one that harnesses the power of citizen participation and blends it with a clever incentive mechanism. **Tom:**  The paper, titled A Participatory Budgeting based Truthful Budget-Limited Incentive Mechanism for Time-Constrained Tasks in Crowdsensing Systems, by Chattu Bhargavi and Vikash Kumar Singh, presents a framework that could revolutionize how we gather data in our cities. **Jen:** It's all about leveraging the collective wisdom and resources of ordinary citizens to address real-world challenges, like environmental monitoring, traffic management, or even identifying areas for urban development. **Tom:** And the best part? The researchers have designed a system that incentivizes participation, ensuring that everyone involved benefits from the process. **Jen:** Intrigued? Let's get into the details!**Body****Jen:** So, imagine a city where residents have a direct say in how public funds are used to collect valuable data. This paper proposes a two-tiered system that does just that. In the first tier, it's all about participatory budgeting.**Tom:** Right. Imagine a town hall meeting where citizens voice their concerns about their community and propose projects they'd like to see funded. In this paper, instead of infrastructure projects, the focus is on data-gathering tasks. **Jen:**  For example, a group of residents might propose a project to monitor air quality in a particular neighborhood. They'd submit a detailed plan, including the tasks involved, the equipment needed, and, most importantly, the estimated budget.**Tom:**  Now, this is where the participatory budgeting aspect gets interesting. The government, with a limited budget, can't fund every proposal. So, they turn to the city dwellers themselves to decide which projects are most worthwhile.**Jen:** Exactly! The city dwellers vote on these proposals, expressing their preferences based on the benefits they perceive for their community.  **Tom:** The researchers designed an algorithm called the Budget Distribution Mechanism to process these votes and select the projects that receive funding. This algorithm works similar to a knapsack problem, where you have to maximize the value of items you put into a knapsack while staying within the weight limit. In this case, the items are the proposals, the value is the benefit to the city, and the weight limit is the government's budget.**Jen:** So, based on the votes, the government allocates funds to the highest-priority projects, effectively creating a pool of funded data collection tasks.**Tom:** But this is just the first tier.  The second tier is where things get truly innovative. Think of it as the execution phase.**Jen:**  Once the projects are chosen, each task needs to be assigned to individuals who can actually carry them out. These are the task executors, and they might be anyone with a smartphone and a willingness to participate.**Tom:**   Let's say the chosen project is to monitor traffic patterns in the city. The task executors would receive instructions on which streets to drive, what data to collect (speed, time, traffic density), and the time frame for completing the task.**Jen:**  Of course, you can't just expect people to do this for free.  Here's where the incentive mechanism comes in.**Tom:**  The paper proposes a Budget-Limited Incentive Mechanism for Crowdsensing (BULINC) that rewards task executors fairly and efficiently.**Jen:**  Think of it like an auction system. Each task executor submits a bid – the minimum amount they’d need to be paid to complete the task. The system then selects the most qualified and cost-effective executors for each task, ensuring that the total payout stays within the allocated budget.**Tom:**  The key here is truthfulness.  The researchers wanted to make sure that the task executors were incentivized to report their true costs, and not to inflate them in order to get a higher payout.**Jen:**  And they designed BULINC to be truthful by using a clever system of marginal pricing. Basically, the system takes into account the costs of the task executors already selected for a task and calculates a fair payment for the next executor. This ensures that no one has an incentive to lie about their costs.**Tom:**  So, not only is the system fair and budget-conscious, but it also encourages honest participation, maximizing the quality and reliability of the data collected.**Jen:**  And the beauty of this system is that it's scalable! It can adapt to projects of varying sizes and complexity, involving more or fewer task executors as needed.**Reflections****Tom:** So, Jen, what did you find most compelling about this research?**Jen:**  I think the most interesting aspect is the idea of combining participatory budgeting with crowdsourcing. It's a powerful combination, creating a truly democratic system where citizens not only decide how data is collected but also benefit directly from the results.**Tom:**  Absolutely. This framework could have a significant impact on urban planning, environmental protection, public health, and even social justice initiatives. It empowers citizens to directly address issues they care about and drive positive change in their communities.**Jen:**  But, of course, there are also some limitations to consider.  The success of this system depends heavily on the level of civic engagement and the availability of individuals willing to participate as task executors.  **Tom:**  And the researchers acknowledge that the framework needs further refinement to address issues like data quality control, privacy protection, and potential biases that could emerge from the voting process.**Jen:**  Nevertheless, this research presents a very promising direction for crowdsourcing and participatory governance, opening the door to a more engaged and data-driven future for our cities.**Conclusion****Tom:**  So, in a nutshell, this paper proposes a two-tiered system that combines participatory budgeting with a truthful incentive mechanism to empower citizens in data collection initiatives. **Jen:**  The research highlights the potential for harnessing the collective wisdom and resources of ordinary people to address real-world challenges, leading to more informed decision-making and a more responsive urban environment. **Tom:**  It's a fascinating example of how technology can be leveraged to foster citizen engagement and create a more sustainable and equitable future for our cities.**Jen:** And that's it for this episode of Research Unraveled. We hope you enjoyed this deep dive into the fascinating world of crowdsourcing and participatory governance. Stay tuned for more insightful explorations of groundbreaking research.  **Tom:** Until next time, keep your minds open and your curiosity alive!\"\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mtts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:29.295646Z\u001b[0m [\u001b[31m\u001b[1merror    \u001b[0m] \u001b[1mException in tts: 400 Either `input.text` or `input.ssml` is longer than the limit of 5000 bytes. This limit is different from quotas. To fix, reduce the byte length of the characters in this request, or consider using the Long Audio API: https://cloud.google.com/text-to-speech/docs/create-audio-text-long-audio-synthesis.\n",
      "NoneType: None\n",
      "\u001b[0m [\u001b[34m\u001b[1mjonlog\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35merror\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:31.363249Z\u001b[0m [\u001b[31m\u001b[1mcritical \u001b[0m] \u001b[1mGoogleTTS error from text=\"## Episode Title: Crowdsourcing the City: How Participatory Budgeting Fuels Citizen-Driven Data Collection**Intro****Tom:** Welcome back to Research Unraveled, the podcast where we take deep dives into the world of cutting-edge research, blending expert analysis with compelling storytelling to illuminate groundbreaking discoveries.**Jen:** And today, we're delving into a fascinating paper that explores a novel approach to crowdsourcing, one that harnesses the power of citizen participation and blends it with a clever incentive mechanism. **Tom:**  The paper, titled A Participatory Budgeting based Truthful Budget-Limited Incentive Mechanism for Time-Constrained Tasks in Crowdsensing Systems, by Chattu Bhargavi and Vikash Kumar Singh, presents a framework that could revolutionize how we gather data in our cities. **Jen:** It's all about leveraging the collective wisdom and resources of ordinary citizens to address real-world challenges, like environmental monitoring, traffic management, or even identifying areas for urban development. **Tom:** And the best part? The researchers have designed a system that incentivizes participation, ensuring that everyone involved benefits from the process. **Jen:** Intrigued? Let's get into the details!**Body****Jen:** So, imagine a city where residents have a direct say in how public funds are used to collect valuable data. This paper proposes a two-tiered system that does just that. In the first tier, it's all about participatory budgeting.**Tom:** Right. Imagine a town hall meeting where citizens voice their concerns about their community and propose projects they'd like to see funded. In this paper, instead of infrastructure projects, the focus is on data-gathering tasks. **Jen:**  For example, a group of residents might propose a project to monitor air quality in a particular neighborhood. They'd submit a detailed plan, including the tasks involved, the equipment needed, and, most importantly, the estimated budget.**Tom:**  Now, this is where the participatory budgeting aspect gets interesting. The government, with a limited budget, can't fund every proposal. So, they turn to the city dwellers themselves to decide which projects are most worthwhile.**Jen:** Exactly! The city dwellers vote on these proposals, expressing their preferences based on the benefits they perceive for their community.  **Tom:** The researchers designed an algorithm called the Budget Distribution Mechanism to process these votes and select the projects that receive funding. This algorithm works similar to a knapsack problem, where you have to maximize the value of items you put into a knapsack while staying within the weight limit. In this case, the items are the proposals, the value is the benefit to the city, and the weight limit is the government's budget.**Jen:** So, based on the votes, the government allocates funds to the highest-priority projects, effectively creating a pool of funded data collection tasks.**Tom:** But this is just the first tier.  The second tier is where things get truly innovative. Think of it as the execution phase.**Jen:**  Once the projects are chosen, each task needs to be assigned to individuals who can actually carry them out. These are the task executors, and they might be anyone with a smartphone and a willingness to participate.**Tom:**   Let's say the chosen project is to monitor traffic patterns in the city. The task executors would receive instructions on which streets to drive, what data to collect (speed, time, traffic density), and the time frame for completing the task.**Jen:**  Of course, you can't just expect people to do this for free.  Here's where the incentive mechanism comes in.**Tom:**  The paper proposes a Budget-Limited Incentive Mechanism for Crowdsensing (BULINC) that rewards task executors fairly and efficiently.**Jen:**  Think of it like an auction system. Each task executor submits a bid – the minimum amount they’d need to be paid to complete the task. The system then selects the most qualified and cost-effective executors for each task, ensuring that the total payout stays within the allocated budget.**Tom:**  The key here is truthfulness.  The researchers wanted to make sure that the task executors were incentivized to report their true costs, and not to inflate them in order to get a higher payout.**Jen:**  And they designed BULINC to be truthful by using a clever system of marginal pricing. Basically, the system takes into account the costs of the task executors already selected for a task and calculates a fair payment for the next executor. This ensures that no one has an incentive to lie about their costs.**Tom:**  So, not only is the system fair and budget-conscious, but it also encourages honest participation, maximizing the quality and reliability of the data collected.**Jen:**  And the beauty of this system is that it's scalable! It can adapt to projects of varying sizes and complexity, involving more or fewer task executors as needed.**Reflections****Tom:** So, Jen, what did you find most compelling about this research?**Jen:**  I think the most interesting aspect is the idea of combining participatory budgeting with crowdsourcing. It's a powerful combination, creating a truly democratic system where citizens not only decide how data is collected but also benefit directly from the results.**Tom:**  Absolutely. This framework could have a significant impact on urban planning, environmental protection, public health, and even social justice initiatives. It empowers citizens to directly address issues they care about and drive positive change in their communities.**Jen:**  But, of course, there are also some limitations to consider.  The success of this system depends heavily on the level of civic engagement and the availability of individuals willing to participate as task executors.  **Tom:**  And the researchers acknowledge that the framework needs further refinement to address issues like data quality control, privacy protection, and potential biases that could emerge from the voting process.**Jen:**  Nevertheless, this research presents a very promising direction for crowdsourcing and participatory governance, opening the door to a more engaged and data-driven future for our cities.**Conclusion****Tom:**  So, in a nutshell, this paper proposes a two-tiered system that combines participatory budgeting with a truthful incentive mechanism to empower citizens in data collection initiatives. **Jen:**  The research highlights the potential for harnessing the collective wisdom and resources of ordinary people to address real-world challenges, leading to more informed decision-making and a more responsive urban environment. **Tom:**  It's a fascinating example of how technology can be leveraged to foster citizen engagement and create a more sustainable and equitable future for our cities.**Jen:** And that's it for this episode of Research Unraveled. We hope you enjoyed this deep dive into the fascinating world of crowdsourcing and participatory governance. Stay tuned for more insightful explorations of groundbreaking research.  **Tom:** Until next time, keep your minds open and your curiosity alive!\"\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mtts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:31.366295Z\u001b[0m [\u001b[31m\u001b[1merror    \u001b[0m] \u001b[1mException in tts: 400 Either `input.text` or `input.ssml` is longer than the limit of 5000 bytes. This limit is different from quotas. To fix, reduce the byte length of the characters in this request, or consider using the Long Audio API: https://cloud.google.com/text-to-speech/docs/create-audio-text-long-audio-synthesis.\n",
      "NoneType: None\n",
      "\u001b[0m [\u001b[34m\u001b[1mjonlog\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35merror\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:33.439176Z\u001b[0m [\u001b[31m\u001b[1mcritical \u001b[0m] \u001b[1mGoogleTTS error from text=\"## Episode Title: Crowdsourcing the City: How Participatory Budgeting Fuels Citizen-Driven Data Collection**Intro****Tom:** Welcome back to Research Unraveled, the podcast where we take deep dives into the world of cutting-edge research, blending expert analysis with compelling storytelling to illuminate groundbreaking discoveries.**Jen:** And today, we're delving into a fascinating paper that explores a novel approach to crowdsourcing, one that harnesses the power of citizen participation and blends it with a clever incentive mechanism. **Tom:**  The paper, titled A Participatory Budgeting based Truthful Budget-Limited Incentive Mechanism for Time-Constrained Tasks in Crowdsensing Systems, by Chattu Bhargavi and Vikash Kumar Singh, presents a framework that could revolutionize how we gather data in our cities. **Jen:** It's all about leveraging the collective wisdom and resources of ordinary citizens to address real-world challenges, like environmental monitoring, traffic management, or even identifying areas for urban development. **Tom:** And the best part? The researchers have designed a system that incentivizes participation, ensuring that everyone involved benefits from the process. **Jen:** Intrigued? Let's get into the details!**Body****Jen:** So, imagine a city where residents have a direct say in how public funds are used to collect valuable data. This paper proposes a two-tiered system that does just that. In the first tier, it's all about participatory budgeting.**Tom:** Right. Imagine a town hall meeting where citizens voice their concerns about their community and propose projects they'd like to see funded. In this paper, instead of infrastructure projects, the focus is on data-gathering tasks. **Jen:**  For example, a group of residents might propose a project to monitor air quality in a particular neighborhood. They'd submit a detailed plan, including the tasks involved, the equipment needed, and, most importantly, the estimated budget.**Tom:**  Now, this is where the participatory budgeting aspect gets interesting. The government, with a limited budget, can't fund every proposal. So, they turn to the city dwellers themselves to decide which projects are most worthwhile.**Jen:** Exactly! The city dwellers vote on these proposals, expressing their preferences based on the benefits they perceive for their community.  **Tom:** The researchers designed an algorithm called the Budget Distribution Mechanism to process these votes and select the projects that receive funding. This algorithm works similar to a knapsack problem, where you have to maximize the value of items you put into a knapsack while staying within the weight limit. In this case, the items are the proposals, the value is the benefit to the city, and the weight limit is the government's budget.**Jen:** So, based on the votes, the government allocates funds to the highest-priority projects, effectively creating a pool of funded data collection tasks.**Tom:** But this is just the first tier.  The second tier is where things get truly innovative. Think of it as the execution phase.**Jen:**  Once the projects are chosen, each task needs to be assigned to individuals who can actually carry them out. These are the task executors, and they might be anyone with a smartphone and a willingness to participate.**Tom:**   Let's say the chosen project is to monitor traffic patterns in the city. The task executors would receive instructions on which streets to drive, what data to collect (speed, time, traffic density), and the time frame for completing the task.**Jen:**  Of course, you can't just expect people to do this for free.  Here's where the incentive mechanism comes in.**Tom:**  The paper proposes a Budget-Limited Incentive Mechanism for Crowdsensing (BULINC) that rewards task executors fairly and efficiently.**Jen:**  Think of it like an auction system. Each task executor submits a bid – the minimum amount they’d need to be paid to complete the task. The system then selects the most qualified and cost-effective executors for each task, ensuring that the total payout stays within the allocated budget.**Tom:**  The key here is truthfulness.  The researchers wanted to make sure that the task executors were incentivized to report their true costs, and not to inflate them in order to get a higher payout.**Jen:**  And they designed BULINC to be truthful by using a clever system of marginal pricing. Basically, the system takes into account the costs of the task executors already selected for a task and calculates a fair payment for the next executor. This ensures that no one has an incentive to lie about their costs.**Tom:**  So, not only is the system fair and budget-conscious, but it also encourages honest participation, maximizing the quality and reliability of the data collected.**Jen:**  And the beauty of this system is that it's scalable! It can adapt to projects of varying sizes and complexity, involving more or fewer task executors as needed.**Reflections****Tom:** So, Jen, what did you find most compelling about this research?**Jen:**  I think the most interesting aspect is the idea of combining participatory budgeting with crowdsourcing. It's a powerful combination, creating a truly democratic system where citizens not only decide how data is collected but also benefit directly from the results.**Tom:**  Absolutely. This framework could have a significant impact on urban planning, environmental protection, public health, and even social justice initiatives. It empowers citizens to directly address issues they care about and drive positive change in their communities.**Jen:**  But, of course, there are also some limitations to consider.  The success of this system depends heavily on the level of civic engagement and the availability of individuals willing to participate as task executors.  **Tom:**  And the researchers acknowledge that the framework needs further refinement to address issues like data quality control, privacy protection, and potential biases that could emerge from the voting process.**Jen:**  Nevertheless, this research presents a very promising direction for crowdsourcing and participatory governance, opening the door to a more engaged and data-driven future for our cities.**Conclusion****Tom:**  So, in a nutshell, this paper proposes a two-tiered system that combines participatory budgeting with a truthful incentive mechanism to empower citizens in data collection initiatives. **Jen:**  The research highlights the potential for harnessing the collective wisdom and resources of ordinary people to address real-world challenges, leading to more informed decision-making and a more responsive urban environment. **Tom:**  It's a fascinating example of how technology can be leveraged to foster citizen engagement and create a more sustainable and equitable future for our cities.**Jen:** And that's it for this episode of Research Unraveled. We hope you enjoyed this deep dive into the fascinating world of crowdsourcing and participatory governance. Stay tuned for more insightful explorations of groundbreaking research.  **Tom:** Until next time, keep your minds open and your curiosity alive!\"\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mtts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:33.441652Z\u001b[0m [\u001b[31m\u001b[1merror    \u001b[0m] \u001b[1mException in tts: 400 Either `input.text` or `input.ssml` is longer than the limit of 5000 bytes. This limit is different from quotas. To fix, reduce the byte length of the characters in this request, or consider using the Long Audio API: https://cloud.google.com/text-to-speech/docs/create-audio-text-long-audio-synthesis.\n",
      "NoneType: None\n",
      "\u001b[0m [\u001b[34m\u001b[1mjonlog\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35merror\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:35.515315Z\u001b[0m [\u001b[31m\u001b[1mcritical \u001b[0m] \u001b[1mGoogleTTS error from text=\"## Episode Title: Crowdsourcing the City: How Participatory Budgeting Fuels Citizen-Driven Data Collection**Intro****Tom:** Welcome back to Research Unraveled, the podcast where we take deep dives into the world of cutting-edge research, blending expert analysis with compelling storytelling to illuminate groundbreaking discoveries.**Jen:** And today, we're delving into a fascinating paper that explores a novel approach to crowdsourcing, one that harnesses the power of citizen participation and blends it with a clever incentive mechanism. **Tom:**  The paper, titled A Participatory Budgeting based Truthful Budget-Limited Incentive Mechanism for Time-Constrained Tasks in Crowdsensing Systems, by Chattu Bhargavi and Vikash Kumar Singh, presents a framework that could revolutionize how we gather data in our cities. **Jen:** It's all about leveraging the collective wisdom and resources of ordinary citizens to address real-world challenges, like environmental monitoring, traffic management, or even identifying areas for urban development. **Tom:** And the best part? The researchers have designed a system that incentivizes participation, ensuring that everyone involved benefits from the process. **Jen:** Intrigued? Let's get into the details!**Body****Jen:** So, imagine a city where residents have a direct say in how public funds are used to collect valuable data. This paper proposes a two-tiered system that does just that. In the first tier, it's all about participatory budgeting.**Tom:** Right. Imagine a town hall meeting where citizens voice their concerns about their community and propose projects they'd like to see funded. In this paper, instead of infrastructure projects, the focus is on data-gathering tasks. **Jen:**  For example, a group of residents might propose a project to monitor air quality in a particular neighborhood. They'd submit a detailed plan, including the tasks involved, the equipment needed, and, most importantly, the estimated budget.**Tom:**  Now, this is where the participatory budgeting aspect gets interesting. The government, with a limited budget, can't fund every proposal. So, they turn to the city dwellers themselves to decide which projects are most worthwhile.**Jen:** Exactly! The city dwellers vote on these proposals, expressing their preferences based on the benefits they perceive for their community.  **Tom:** The researchers designed an algorithm called the Budget Distribution Mechanism to process these votes and select the projects that receive funding. This algorithm works similar to a knapsack problem, where you have to maximize the value of items you put into a knapsack while staying within the weight limit. In this case, the items are the proposals, the value is the benefit to the city, and the weight limit is the government's budget.**Jen:** So, based on the votes, the government allocates funds to the highest-priority projects, effectively creating a pool of funded data collection tasks.**Tom:** But this is just the first tier.  The second tier is where things get truly innovative. Think of it as the execution phase.**Jen:**  Once the projects are chosen, each task needs to be assigned to individuals who can actually carry them out. These are the task executors, and they might be anyone with a smartphone and a willingness to participate.**Tom:**   Let's say the chosen project is to monitor traffic patterns in the city. The task executors would receive instructions on which streets to drive, what data to collect (speed, time, traffic density), and the time frame for completing the task.**Jen:**  Of course, you can't just expect people to do this for free.  Here's where the incentive mechanism comes in.**Tom:**  The paper proposes a Budget-Limited Incentive Mechanism for Crowdsensing (BULINC) that rewards task executors fairly and efficiently.**Jen:**  Think of it like an auction system. Each task executor submits a bid – the minimum amount they’d need to be paid to complete the task. The system then selects the most qualified and cost-effective executors for each task, ensuring that the total payout stays within the allocated budget.**Tom:**  The key here is truthfulness.  The researchers wanted to make sure that the task executors were incentivized to report their true costs, and not to inflate them in order to get a higher payout.**Jen:**  And they designed BULINC to be truthful by using a clever system of marginal pricing. Basically, the system takes into account the costs of the task executors already selected for a task and calculates a fair payment for the next executor. This ensures that no one has an incentive to lie about their costs.**Tom:**  So, not only is the system fair and budget-conscious, but it also encourages honest participation, maximizing the quality and reliability of the data collected.**Jen:**  And the beauty of this system is that it's scalable! It can adapt to projects of varying sizes and complexity, involving more or fewer task executors as needed.**Reflections****Tom:** So, Jen, what did you find most compelling about this research?**Jen:**  I think the most interesting aspect is the idea of combining participatory budgeting with crowdsourcing. It's a powerful combination, creating a truly democratic system where citizens not only decide how data is collected but also benefit directly from the results.**Tom:**  Absolutely. This framework could have a significant impact on urban planning, environmental protection, public health, and even social justice initiatives. It empowers citizens to directly address issues they care about and drive positive change in their communities.**Jen:**  But, of course, there are also some limitations to consider.  The success of this system depends heavily on the level of civic engagement and the availability of individuals willing to participate as task executors.  **Tom:**  And the researchers acknowledge that the framework needs further refinement to address issues like data quality control, privacy protection, and potential biases that could emerge from the voting process.**Jen:**  Nevertheless, this research presents a very promising direction for crowdsourcing and participatory governance, opening the door to a more engaged and data-driven future for our cities.**Conclusion****Tom:**  So, in a nutshell, this paper proposes a two-tiered system that combines participatory budgeting with a truthful incentive mechanism to empower citizens in data collection initiatives. **Jen:**  The research highlights the potential for harnessing the collective wisdom and resources of ordinary people to address real-world challenges, leading to more informed decision-making and a more responsive urban environment. **Tom:**  It's a fascinating example of how technology can be leveraged to foster citizen engagement and create a more sustainable and equitable future for our cities.**Jen:** And that's it for this episode of Research Unraveled. We hope you enjoyed this deep dive into the fascinating world of crowdsourcing and participatory governance. Stay tuned for more insightful explorations of groundbreaking research.  **Tom:** Until next time, keep your minds open and your curiosity alive!\"\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mtts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:35.518082Z\u001b[0m [\u001b[31m\u001b[1merror    \u001b[0m] \u001b[1mException in tts: 400 Either `input.text` or `input.ssml` is longer than the limit of 5000 bytes. This limit is different from quotas. To fix, reduce the byte length of the characters in this request, or consider using the Long Audio API: https://cloud.google.com/text-to-speech/docs/create-audio-text-long-audio-synthesis.\n",
      "NoneType: None\n",
      "\u001b[0m [\u001b[34m\u001b[1mjonlog\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35merror\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:37.661958Z\u001b[0m [\u001b[31m\u001b[1mcritical \u001b[0m] \u001b[1mGoogleTTS error from text=\"## Episode Title: Crowdsourcing the City: How Participatory Budgeting Fuels Citizen-Driven Data Collection**Intro****Tom:** Welcome back to Research Unraveled, the podcast where we take deep dives into the world of cutting-edge research, blending expert analysis with compelling storytelling to illuminate groundbreaking discoveries.**Jen:** And today, we're delving into a fascinating paper that explores a novel approach to crowdsourcing, one that harnesses the power of citizen participation and blends it with a clever incentive mechanism. **Tom:**  The paper, titled A Participatory Budgeting based Truthful Budget-Limited Incentive Mechanism for Time-Constrained Tasks in Crowdsensing Systems, by Chattu Bhargavi and Vikash Kumar Singh, presents a framework that could revolutionize how we gather data in our cities. **Jen:** It's all about leveraging the collective wisdom and resources of ordinary citizens to address real-world challenges, like environmental monitoring, traffic management, or even identifying areas for urban development. **Tom:** And the best part? The researchers have designed a system that incentivizes participation, ensuring that everyone involved benefits from the process. **Jen:** Intrigued? Let's get into the details!**Body****Jen:** So, imagine a city where residents have a direct say in how public funds are used to collect valuable data. This paper proposes a two-tiered system that does just that. In the first tier, it's all about participatory budgeting.**Tom:** Right. Imagine a town hall meeting where citizens voice their concerns about their community and propose projects they'd like to see funded. In this paper, instead of infrastructure projects, the focus is on data-gathering tasks. **Jen:**  For example, a group of residents might propose a project to monitor air quality in a particular neighborhood. They'd submit a detailed plan, including the tasks involved, the equipment needed, and, most importantly, the estimated budget.**Tom:**  Now, this is where the participatory budgeting aspect gets interesting. The government, with a limited budget, can't fund every proposal. So, they turn to the city dwellers themselves to decide which projects are most worthwhile.**Jen:** Exactly! The city dwellers vote on these proposals, expressing their preferences based on the benefits they perceive for their community.  **Tom:** The researchers designed an algorithm called the Budget Distribution Mechanism to process these votes and select the projects that receive funding. This algorithm works similar to a knapsack problem, where you have to maximize the value of items you put into a knapsack while staying within the weight limit. In this case, the items are the proposals, the value is the benefit to the city, and the weight limit is the government's budget.**Jen:** So, based on the votes, the government allocates funds to the highest-priority projects, effectively creating a pool of funded data collection tasks.**Tom:** But this is just the first tier.  The second tier is where things get truly innovative. Think of it as the execution phase.**Jen:**  Once the projects are chosen, each task needs to be assigned to individuals who can actually carry them out. These are the task executors, and they might be anyone with a smartphone and a willingness to participate.**Tom:**   Let's say the chosen project is to monitor traffic patterns in the city. The task executors would receive instructions on which streets to drive, what data to collect (speed, time, traffic density), and the time frame for completing the task.**Jen:**  Of course, you can't just expect people to do this for free.  Here's where the incentive mechanism comes in.**Tom:**  The paper proposes a Budget-Limited Incentive Mechanism for Crowdsensing (BULINC) that rewards task executors fairly and efficiently.**Jen:**  Think of it like an auction system. Each task executor submits a bid – the minimum amount they’d need to be paid to complete the task. The system then selects the most qualified and cost-effective executors for each task, ensuring that the total payout stays within the allocated budget.**Tom:**  The key here is truthfulness.  The researchers wanted to make sure that the task executors were incentivized to report their true costs, and not to inflate them in order to get a higher payout.**Jen:**  And they designed BULINC to be truthful by using a clever system of marginal pricing. Basically, the system takes into account the costs of the task executors already selected for a task and calculates a fair payment for the next executor. This ensures that no one has an incentive to lie about their costs.**Tom:**  So, not only is the system fair and budget-conscious, but it also encourages honest participation, maximizing the quality and reliability of the data collected.**Jen:**  And the beauty of this system is that it's scalable! It can adapt to projects of varying sizes and complexity, involving more or fewer task executors as needed.**Reflections****Tom:** So, Jen, what did you find most compelling about this research?**Jen:**  I think the most interesting aspect is the idea of combining participatory budgeting with crowdsourcing. It's a powerful combination, creating a truly democratic system where citizens not only decide how data is collected but also benefit directly from the results.**Tom:**  Absolutely. This framework could have a significant impact on urban planning, environmental protection, public health, and even social justice initiatives. It empowers citizens to directly address issues they care about and drive positive change in their communities.**Jen:**  But, of course, there are also some limitations to consider.  The success of this system depends heavily on the level of civic engagement and the availability of individuals willing to participate as task executors.  **Tom:**  And the researchers acknowledge that the framework needs further refinement to address issues like data quality control, privacy protection, and potential biases that could emerge from the voting process.**Jen:**  Nevertheless, this research presents a very promising direction for crowdsourcing and participatory governance, opening the door to a more engaged and data-driven future for our cities.**Conclusion****Tom:**  So, in a nutshell, this paper proposes a two-tiered system that combines participatory budgeting with a truthful incentive mechanism to empower citizens in data collection initiatives. **Jen:**  The research highlights the potential for harnessing the collective wisdom and resources of ordinary people to address real-world challenges, leading to more informed decision-making and a more responsive urban environment. **Tom:**  It's a fascinating example of how technology can be leveraged to foster citizen engagement and create a more sustainable and equitable future for our cities.**Jen:** And that's it for this episode of Research Unraveled. We hope you enjoyed this deep dive into the fascinating world of crowdsourcing and participatory governance. Stay tuned for more insightful explorations of groundbreaking research.  **Tom:** Until next time, keep your minds open and your curiosity alive!\"\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mtts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:37.665241Z\u001b[0m [\u001b[31m\u001b[1merror    \u001b[0m] \u001b[1mException in tts: 400 Either `input.text` or `input.ssml` is longer than the limit of 5000 bytes. This limit is different from quotas. To fix, reduce the byte length of the characters in this request, or consider using the Long Audio API: https://cloud.google.com/text-to-speech/docs/create-audio-text-long-audio-synthesis.\n",
      "NoneType: None\n",
      "\u001b[0m [\u001b[34m\u001b[1mjonlog\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35merror\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-5_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:39.725352Z\u001b[0m [\u001b[32m\u001b[1minfo     \u001b[0m] \u001b[1mCurrently at self.num_tokens_from_messages(self._history)=21114 tokens in conversation\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mmessage\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145707151360\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-4_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:57:39.726339Z\u001b[0m [\u001b[32m\u001b[1minfo     \u001b[0m] \u001b[1mrequesting chatcompletion model='GOOGLE/gemini-1.5-flash-latest'...\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35m_msg\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145707151360\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-4_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:01.219427Z\u001b[0m [\u001b[32m\u001b[1minfo     \u001b[0m] \u001b[1mreceived chatcompletion model='gemini-1.5-flash-latest'...\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35m_msg\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145707151360\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-4_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:01.222342Z\u001b[0m [\u001b[32m\u001b[1minfo     \u001b[0m] \u001b[1mrequesting tts i=0 voice=[GoogleTTS] self._voice_name='en-US-Wavenet-J'\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mwrite_audio\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:01.343748Z\u001b[0m [\u001b[31m\u001b[1mcritical \u001b[0m] \u001b[1mGoogleTTS error from text=\"## Episode Title: Crowdsourcing, Democracy, and Dollars: How to Get the Most Out of Your City's Data**Intro****Tom:** Welcome back to Research Unraveled, the podcast where we take complex research papers and break them down for everyone to understand. I'm your host, Tom, and with me, as always, is the amazing Jen. **Jen:** Hey everyone! Today we're diving into a fascinating paper that examines how to make crowdsensing, the practice of gathering data from everyday people, more efficient and fair.  **Tom:**  This isn't just about collecting random data, Jen.  We're talking about harnessing the power of the crowd to tackle real-world problems, like monitoring environmental changes or traffic congestion, and they do it using a system that combines participatory budgeting with incentives. **Jen:** Intriguing! It sounds like this paper is merging technology and democracy in a really unique way.  **Tom:**  It definitely is. The title of the paper is A Participatory Budgeting based Truthful Budget-Limited Incentive Mechanism for Time-Constrained Tasks in Crowdsensing Systems. Quite a mouthful, right? But don't worry, we'll break it all down. **Body****Jen:**  Okay, let's start at the beginning. What's crowdsensing, and why is it so important?**Tom:**  Imagine you want to gather data about traffic patterns in your city, but you don't have the resources to install sensors everywhere.  Crowdsensing offers a solution.  It leverages the fact that people carry smartphones and other devices that can be used as sensors. They can be the eyes and ears of your city, contributing valuable data.**Jen:**  That makes sense! But how do you convince people to participate?  That's where the incentive mechanism comes in, right?**Tom:**  Exactly. The authors of this paper propose a system where the government, or any funding organization, provides a limited budget to support these crowdsensing projects.  This budget is allocated based on a participatory budgeting approach.**Jen:**  What's participatory budgeting?  **Tom:**  Instead of the government deciding which projects get funded, the local citizens get to vote. They express their preferences for projects they think are most important. Think about it like a city-wide poll, where people can weigh in on whether they want more streetlights, better parks, or improved public transportation.**Jen:**  Wow, so it's like a direct democracy for crowdsensing!  That's empowering.  But how do you ensure that the budget is used effectively and that the people contributing data are fairly compensated?**Tom:** This is where things get interesting. The authors have devised a two-tiered system.  **Jen:**  Two tiers?**Tom:**  Yes. First, the participatory budgeting process determines which crowdsensing projects get funding. This involves the city dwellers voting on their preferred projects, and then the government allocates its limited funds based on the votes. **Jen:**  So, it's like a popularity contest where the projects with the most votes get funded?**Tom:**  Not exactly. They have an algorithm to optimize the budget allocation. The algorithm aims to ensure that the projects with the most potential benefit to the city, as perceived by the citizens, get prioritized.  It's a more sophisticated system that takes into account the estimated costs of each project and the total available budget.**Jen:**  Okay, that makes sense.  But what about the second tier?**Tom:**  The second tier is about actually executing the crowdsensing projects.  Once a project is funded, it needs to be divided into tasks that can be performed by the crowd, and those tasks need to be scheduled in a way that doesn't create conflicts.  **Jen:**  Like, you wouldn't want two people trying to collect traffic data at the same spot at the same time, right?**Tom:**  Exactly. They use a scheduling algorithm to make sure that the tasks are assigned to different time slots, preventing overlap and ensuring that all the data is collected efficiently. **Jen:**  So far, it seems like a very well-thought-out system! But I'm curious, how are the people participating in the data collection actually rewarded?**Tom:**  That's where the budget-limited incentive mechanism comes in. Each crowdsensing task has a set budget assigned to it. The people participating, who we call task executors, submit bids for how much they're willing to do the task for. The algorithm then selects the executors who offer the lowest bids, making sure the total cost stays within the budget.**Jen:**  It's like an auction, but with a fixed budget? **Tom:**  You could say that.  And here's the key part – the authors have designed the system to be truthful.**Jen:**  What does that mean?**Tom:**  It means that the task executors have no incentive to lie about their bids.  If they try to inflate their bids to get a higher payment, the algorithm might not choose them, and they'll get nothing.  **Jen:**  So, honesty is the best policy!  That's important for ensuring fairness. **Tom:**  Absolutely!  The authors have done a lot of theoretical analysis to prove that this system is fair and efficient. They've shown that it's computationally efficient, meaning it doesn't take too long to run, and it's individually rational, meaning that the task executors will always make a profit or at least break even.  **Jen:**  This all sounds fantastic! But I'm wondering, how do the researchers know this actually works in the real world?**Tom:**  Good question! They've conducted simulations using real-world data from participatory democracy initiatives.  **Jen:**  So, they're not just theorizing!  **Tom:**  Right!  The simulation results support their conclusions. They found that, as the probability of a project getting funded increases, the number of projects that actually get funded also increases.  **Jen:**  That makes sense.  The more likely people are to vote for a project, the more likely it is to get the funds.**Tom:**  And they also tested the incentive mechanism by comparing their system to other existing mechanisms.  They found that their system consistently outperformed the others in terms of maximizing the overall profit of the task executors, while still staying within the budget constraints.**Reflections****Jen:**  This paper really brings together a lot of interesting elements – democracy, crowdsourcing, and even a bit of game theory to ensure fairness. **Tom:**  I agree.  It's a great example of how technology can be used to empower citizens and create more efficient systems. **Jen:**  I'm also impressed by the authors' thorough analysis. They didn't just throw ideas out there; they actually proved that their system works.**Tom:**  Definitely.  And the fact that they used real-world data adds even more weight to their findings.  **Jen:**  However, I do have a question.  The paper focuses on time-constrained tasks.  What about crowdsensing projects that are not limited by time?  Would this system still work? **Tom:**  That's a great point, Jen.  This paper lays the groundwork, but there's definitely room for future research to explore how this system could be adapted to different types of crowdsensing projects. **Jen:**  And what about the potential for people to manipulate the voting process?  **Tom:**  Another great point!  The authors acknowledge that this is a possible concern and suggest further research to explore ways to mitigate it.  **Conclusion****Tom:**  So, to wrap things up, we've learned about a clever system for incentivizing participation in crowdsensing projects, using a combination of participatory budgeting and a truthful incentive mechanism.  **Jen:**  This research shows the potential for merging technology and democracy to create more efficient and equitable systems.  It's inspiring to see how we can leverage the collective wisdom of the crowd to solve real-world problems.**Tom:**  And it's just the beginning.  As we move forward, we can expect to see even more innovative approaches to crowdsensing, fueled by the principles of transparency, fairness, and citizen empowerment.**Jen:** Thanks for joining us today, and remember, stay curious! We'll see you in the next episode of Research Unraveled.\"\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mtts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:01.346714Z\u001b[0m [\u001b[31m\u001b[1merror    \u001b[0m] \u001b[1mException in tts: 400 Either `input.text` or `input.ssml` is longer than the limit of 5000 bytes. This limit is different from quotas. To fix, reduce the byte length of the characters in this request, or consider using the Long Audio API: https://cloud.google.com/text-to-speech/docs/create-audio-text-long-audio-synthesis.\n",
      "NoneType: None\n",
      "\u001b[0m [\u001b[34m\u001b[1mjonlog\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35merror\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:03.394272Z\u001b[0m [\u001b[31m\u001b[1mcritical \u001b[0m] \u001b[1mGoogleTTS error from text=\"## Episode Title: Crowdsourcing, Democracy, and Dollars: How to Get the Most Out of Your City's Data**Intro****Tom:** Welcome back to Research Unraveled, the podcast where we take complex research papers and break them down for everyone to understand. I'm your host, Tom, and with me, as always, is the amazing Jen. **Jen:** Hey everyone! Today we're diving into a fascinating paper that examines how to make crowdsensing, the practice of gathering data from everyday people, more efficient and fair.  **Tom:**  This isn't just about collecting random data, Jen.  We're talking about harnessing the power of the crowd to tackle real-world problems, like monitoring environmental changes or traffic congestion, and they do it using a system that combines participatory budgeting with incentives. **Jen:** Intriguing! It sounds like this paper is merging technology and democracy in a really unique way.  **Tom:**  It definitely is. The title of the paper is A Participatory Budgeting based Truthful Budget-Limited Incentive Mechanism for Time-Constrained Tasks in Crowdsensing Systems. Quite a mouthful, right? But don't worry, we'll break it all down. **Body****Jen:**  Okay, let's start at the beginning. What's crowdsensing, and why is it so important?**Tom:**  Imagine you want to gather data about traffic patterns in your city, but you don't have the resources to install sensors everywhere.  Crowdsensing offers a solution.  It leverages the fact that people carry smartphones and other devices that can be used as sensors. They can be the eyes and ears of your city, contributing valuable data.**Jen:**  That makes sense! But how do you convince people to participate?  That's where the incentive mechanism comes in, right?**Tom:**  Exactly. The authors of this paper propose a system where the government, or any funding organization, provides a limited budget to support these crowdsensing projects.  This budget is allocated based on a participatory budgeting approach.**Jen:**  What's participatory budgeting?  **Tom:**  Instead of the government deciding which projects get funded, the local citizens get to vote. They express their preferences for projects they think are most important. Think about it like a city-wide poll, where people can weigh in on whether they want more streetlights, better parks, or improved public transportation.**Jen:**  Wow, so it's like a direct democracy for crowdsensing!  That's empowering.  But how do you ensure that the budget is used effectively and that the people contributing data are fairly compensated?**Tom:** This is where things get interesting. The authors have devised a two-tiered system.  **Jen:**  Two tiers?**Tom:**  Yes. First, the participatory budgeting process determines which crowdsensing projects get funding. This involves the city dwellers voting on their preferred projects, and then the government allocates its limited funds based on the votes. **Jen:**  So, it's like a popularity contest where the projects with the most votes get funded?**Tom:**  Not exactly. They have an algorithm to optimize the budget allocation. The algorithm aims to ensure that the projects with the most potential benefit to the city, as perceived by the citizens, get prioritized.  It's a more sophisticated system that takes into account the estimated costs of each project and the total available budget.**Jen:**  Okay, that makes sense.  But what about the second tier?**Tom:**  The second tier is about actually executing the crowdsensing projects.  Once a project is funded, it needs to be divided into tasks that can be performed by the crowd, and those tasks need to be scheduled in a way that doesn't create conflicts.  **Jen:**  Like, you wouldn't want two people trying to collect traffic data at the same spot at the same time, right?**Tom:**  Exactly. They use a scheduling algorithm to make sure that the tasks are assigned to different time slots, preventing overlap and ensuring that all the data is collected efficiently. **Jen:**  So far, it seems like a very well-thought-out system! But I'm curious, how are the people participating in the data collection actually rewarded?**Tom:**  That's where the budget-limited incentive mechanism comes in. Each crowdsensing task has a set budget assigned to it. The people participating, who we call task executors, submit bids for how much they're willing to do the task for. The algorithm then selects the executors who offer the lowest bids, making sure the total cost stays within the budget.**Jen:**  It's like an auction, but with a fixed budget? **Tom:**  You could say that.  And here's the key part – the authors have designed the system to be truthful.**Jen:**  What does that mean?**Tom:**  It means that the task executors have no incentive to lie about their bids.  If they try to inflate their bids to get a higher payment, the algorithm might not choose them, and they'll get nothing.  **Jen:**  So, honesty is the best policy!  That's important for ensuring fairness. **Tom:**  Absolutely!  The authors have done a lot of theoretical analysis to prove that this system is fair and efficient. They've shown that it's computationally efficient, meaning it doesn't take too long to run, and it's individually rational, meaning that the task executors will always make a profit or at least break even.  **Jen:**  This all sounds fantastic! But I'm wondering, how do the researchers know this actually works in the real world?**Tom:**  Good question! They've conducted simulations using real-world data from participatory democracy initiatives.  **Jen:**  So, they're not just theorizing!  **Tom:**  Right!  The simulation results support their conclusions. They found that, as the probability of a project getting funded increases, the number of projects that actually get funded also increases.  **Jen:**  That makes sense.  The more likely people are to vote for a project, the more likely it is to get the funds.**Tom:**  And they also tested the incentive mechanism by comparing their system to other existing mechanisms.  They found that their system consistently outperformed the others in terms of maximizing the overall profit of the task executors, while still staying within the budget constraints.**Reflections****Jen:**  This paper really brings together a lot of interesting elements – democracy, crowdsourcing, and even a bit of game theory to ensure fairness. **Tom:**  I agree.  It's a great example of how technology can be used to empower citizens and create more efficient systems. **Jen:**  I'm also impressed by the authors' thorough analysis. They didn't just throw ideas out there; they actually proved that their system works.**Tom:**  Definitely.  And the fact that they used real-world data adds even more weight to their findings.  **Jen:**  However, I do have a question.  The paper focuses on time-constrained tasks.  What about crowdsensing projects that are not limited by time?  Would this system still work? **Tom:**  That's a great point, Jen.  This paper lays the groundwork, but there's definitely room for future research to explore how this system could be adapted to different types of crowdsensing projects. **Jen:**  And what about the potential for people to manipulate the voting process?  **Tom:**  Another great point!  The authors acknowledge that this is a possible concern and suggest further research to explore ways to mitigate it.  **Conclusion****Tom:**  So, to wrap things up, we've learned about a clever system for incentivizing participation in crowdsensing projects, using a combination of participatory budgeting and a truthful incentive mechanism.  **Jen:**  This research shows the potential for merging technology and democracy to create more efficient and equitable systems.  It's inspiring to see how we can leverage the collective wisdom of the crowd to solve real-world problems.**Tom:**  And it's just the beginning.  As we move forward, we can expect to see even more innovative approaches to crowdsensing, fueled by the principles of transparency, fairness, and citizen empowerment.**Jen:** Thanks for joining us today, and remember, stay curious! We'll see you in the next episode of Research Unraveled.\"\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mtts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:03.397404Z\u001b[0m [\u001b[31m\u001b[1merror    \u001b[0m] \u001b[1mException in tts: 400 Either `input.text` or `input.ssml` is longer than the limit of 5000 bytes. This limit is different from quotas. To fix, reduce the byte length of the characters in this request, or consider using the Long Audio API: https://cloud.google.com/text-to-speech/docs/create-audio-text-long-audio-synthesis.\n",
      "NoneType: None\n",
      "\u001b[0m [\u001b[34m\u001b[1mjonlog\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35merror\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:05.445865Z\u001b[0m [\u001b[31m\u001b[1mcritical \u001b[0m] \u001b[1mGoogleTTS error from text=\"## Episode Title: Crowdsourcing, Democracy, and Dollars: How to Get the Most Out of Your City's Data**Intro****Tom:** Welcome back to Research Unraveled, the podcast where we take complex research papers and break them down for everyone to understand. I'm your host, Tom, and with me, as always, is the amazing Jen. **Jen:** Hey everyone! Today we're diving into a fascinating paper that examines how to make crowdsensing, the practice of gathering data from everyday people, more efficient and fair.  **Tom:**  This isn't just about collecting random data, Jen.  We're talking about harnessing the power of the crowd to tackle real-world problems, like monitoring environmental changes or traffic congestion, and they do it using a system that combines participatory budgeting with incentives. **Jen:** Intriguing! It sounds like this paper is merging technology and democracy in a really unique way.  **Tom:**  It definitely is. The title of the paper is A Participatory Budgeting based Truthful Budget-Limited Incentive Mechanism for Time-Constrained Tasks in Crowdsensing Systems. Quite a mouthful, right? But don't worry, we'll break it all down. **Body****Jen:**  Okay, let's start at the beginning. What's crowdsensing, and why is it so important?**Tom:**  Imagine you want to gather data about traffic patterns in your city, but you don't have the resources to install sensors everywhere.  Crowdsensing offers a solution.  It leverages the fact that people carry smartphones and other devices that can be used as sensors. They can be the eyes and ears of your city, contributing valuable data.**Jen:**  That makes sense! But how do you convince people to participate?  That's where the incentive mechanism comes in, right?**Tom:**  Exactly. The authors of this paper propose a system where the government, or any funding organization, provides a limited budget to support these crowdsensing projects.  This budget is allocated based on a participatory budgeting approach.**Jen:**  What's participatory budgeting?  **Tom:**  Instead of the government deciding which projects get funded, the local citizens get to vote. They express their preferences for projects they think are most important. Think about it like a city-wide poll, where people can weigh in on whether they want more streetlights, better parks, or improved public transportation.**Jen:**  Wow, so it's like a direct democracy for crowdsensing!  That's empowering.  But how do you ensure that the budget is used effectively and that the people contributing data are fairly compensated?**Tom:** This is where things get interesting. The authors have devised a two-tiered system.  **Jen:**  Two tiers?**Tom:**  Yes. First, the participatory budgeting process determines which crowdsensing projects get funding. This involves the city dwellers voting on their preferred projects, and then the government allocates its limited funds based on the votes. **Jen:**  So, it's like a popularity contest where the projects with the most votes get funded?**Tom:**  Not exactly. They have an algorithm to optimize the budget allocation. The algorithm aims to ensure that the projects with the most potential benefit to the city, as perceived by the citizens, get prioritized.  It's a more sophisticated system that takes into account the estimated costs of each project and the total available budget.**Jen:**  Okay, that makes sense.  But what about the second tier?**Tom:**  The second tier is about actually executing the crowdsensing projects.  Once a project is funded, it needs to be divided into tasks that can be performed by the crowd, and those tasks need to be scheduled in a way that doesn't create conflicts.  **Jen:**  Like, you wouldn't want two people trying to collect traffic data at the same spot at the same time, right?**Tom:**  Exactly. They use a scheduling algorithm to make sure that the tasks are assigned to different time slots, preventing overlap and ensuring that all the data is collected efficiently. **Jen:**  So far, it seems like a very well-thought-out system! But I'm curious, how are the people participating in the data collection actually rewarded?**Tom:**  That's where the budget-limited incentive mechanism comes in. Each crowdsensing task has a set budget assigned to it. The people participating, who we call task executors, submit bids for how much they're willing to do the task for. The algorithm then selects the executors who offer the lowest bids, making sure the total cost stays within the budget.**Jen:**  It's like an auction, but with a fixed budget? **Tom:**  You could say that.  And here's the key part – the authors have designed the system to be truthful.**Jen:**  What does that mean?**Tom:**  It means that the task executors have no incentive to lie about their bids.  If they try to inflate their bids to get a higher payment, the algorithm might not choose them, and they'll get nothing.  **Jen:**  So, honesty is the best policy!  That's important for ensuring fairness. **Tom:**  Absolutely!  The authors have done a lot of theoretical analysis to prove that this system is fair and efficient. They've shown that it's computationally efficient, meaning it doesn't take too long to run, and it's individually rational, meaning that the task executors will always make a profit or at least break even.  **Jen:**  This all sounds fantastic! But I'm wondering, how do the researchers know this actually works in the real world?**Tom:**  Good question! They've conducted simulations using real-world data from participatory democracy initiatives.  **Jen:**  So, they're not just theorizing!  **Tom:**  Right!  The simulation results support their conclusions. They found that, as the probability of a project getting funded increases, the number of projects that actually get funded also increases.  **Jen:**  That makes sense.  The more likely people are to vote for a project, the more likely it is to get the funds.**Tom:**  And they also tested the incentive mechanism by comparing their system to other existing mechanisms.  They found that their system consistently outperformed the others in terms of maximizing the overall profit of the task executors, while still staying within the budget constraints.**Reflections****Jen:**  This paper really brings together a lot of interesting elements – democracy, crowdsourcing, and even a bit of game theory to ensure fairness. **Tom:**  I agree.  It's a great example of how technology can be used to empower citizens and create more efficient systems. **Jen:**  I'm also impressed by the authors' thorough analysis. They didn't just throw ideas out there; they actually proved that their system works.**Tom:**  Definitely.  And the fact that they used real-world data adds even more weight to their findings.  **Jen:**  However, I do have a question.  The paper focuses on time-constrained tasks.  What about crowdsensing projects that are not limited by time?  Would this system still work? **Tom:**  That's a great point, Jen.  This paper lays the groundwork, but there's definitely room for future research to explore how this system could be adapted to different types of crowdsensing projects. **Jen:**  And what about the potential for people to manipulate the voting process?  **Tom:**  Another great point!  The authors acknowledge that this is a possible concern and suggest further research to explore ways to mitigate it.  **Conclusion****Tom:**  So, to wrap things up, we've learned about a clever system for incentivizing participation in crowdsensing projects, using a combination of participatory budgeting and a truthful incentive mechanism.  **Jen:**  This research shows the potential for merging technology and democracy to create more efficient and equitable systems.  It's inspiring to see how we can leverage the collective wisdom of the crowd to solve real-world problems.**Tom:**  And it's just the beginning.  As we move forward, we can expect to see even more innovative approaches to crowdsensing, fueled by the principles of transparency, fairness, and citizen empowerment.**Jen:** Thanks for joining us today, and remember, stay curious! We'll see you in the next episode of Research Unraveled.\"\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mtts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:05.447517Z\u001b[0m [\u001b[31m\u001b[1merror    \u001b[0m] \u001b[1mException in tts: 400 Either `input.text` or `input.ssml` is longer than the limit of 5000 bytes. This limit is different from quotas. To fix, reduce the byte length of the characters in this request, or consider using the Long Audio API: https://cloud.google.com/text-to-speech/docs/create-audio-text-long-audio-synthesis.\n",
      "NoneType: None\n",
      "\u001b[0m [\u001b[34m\u001b[1mjonlog\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35merror\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:07.522481Z\u001b[0m [\u001b[31m\u001b[1mcritical \u001b[0m] \u001b[1mGoogleTTS error from text=\"## Episode Title: Crowdsourcing, Democracy, and Dollars: How to Get the Most Out of Your City's Data**Intro****Tom:** Welcome back to Research Unraveled, the podcast where we take complex research papers and break them down for everyone to understand. I'm your host, Tom, and with me, as always, is the amazing Jen. **Jen:** Hey everyone! Today we're diving into a fascinating paper that examines how to make crowdsensing, the practice of gathering data from everyday people, more efficient and fair.  **Tom:**  This isn't just about collecting random data, Jen.  We're talking about harnessing the power of the crowd to tackle real-world problems, like monitoring environmental changes or traffic congestion, and they do it using a system that combines participatory budgeting with incentives. **Jen:** Intriguing! It sounds like this paper is merging technology and democracy in a really unique way.  **Tom:**  It definitely is. The title of the paper is A Participatory Budgeting based Truthful Budget-Limited Incentive Mechanism for Time-Constrained Tasks in Crowdsensing Systems. Quite a mouthful, right? But don't worry, we'll break it all down. **Body****Jen:**  Okay, let's start at the beginning. What's crowdsensing, and why is it so important?**Tom:**  Imagine you want to gather data about traffic patterns in your city, but you don't have the resources to install sensors everywhere.  Crowdsensing offers a solution.  It leverages the fact that people carry smartphones and other devices that can be used as sensors. They can be the eyes and ears of your city, contributing valuable data.**Jen:**  That makes sense! But how do you convince people to participate?  That's where the incentive mechanism comes in, right?**Tom:**  Exactly. The authors of this paper propose a system where the government, or any funding organization, provides a limited budget to support these crowdsensing projects.  This budget is allocated based on a participatory budgeting approach.**Jen:**  What's participatory budgeting?  **Tom:**  Instead of the government deciding which projects get funded, the local citizens get to vote. They express their preferences for projects they think are most important. Think about it like a city-wide poll, where people can weigh in on whether they want more streetlights, better parks, or improved public transportation.**Jen:**  Wow, so it's like a direct democracy for crowdsensing!  That's empowering.  But how do you ensure that the budget is used effectively and that the people contributing data are fairly compensated?**Tom:** This is where things get interesting. The authors have devised a two-tiered system.  **Jen:**  Two tiers?**Tom:**  Yes. First, the participatory budgeting process determines which crowdsensing projects get funding. This involves the city dwellers voting on their preferred projects, and then the government allocates its limited funds based on the votes. **Jen:**  So, it's like a popularity contest where the projects with the most votes get funded?**Tom:**  Not exactly. They have an algorithm to optimize the budget allocation. The algorithm aims to ensure that the projects with the most potential benefit to the city, as perceived by the citizens, get prioritized.  It's a more sophisticated system that takes into account the estimated costs of each project and the total available budget.**Jen:**  Okay, that makes sense.  But what about the second tier?**Tom:**  The second tier is about actually executing the crowdsensing projects.  Once a project is funded, it needs to be divided into tasks that can be performed by the crowd, and those tasks need to be scheduled in a way that doesn't create conflicts.  **Jen:**  Like, you wouldn't want two people trying to collect traffic data at the same spot at the same time, right?**Tom:**  Exactly. They use a scheduling algorithm to make sure that the tasks are assigned to different time slots, preventing overlap and ensuring that all the data is collected efficiently. **Jen:**  So far, it seems like a very well-thought-out system! But I'm curious, how are the people participating in the data collection actually rewarded?**Tom:**  That's where the budget-limited incentive mechanism comes in. Each crowdsensing task has a set budget assigned to it. The people participating, who we call task executors, submit bids for how much they're willing to do the task for. The algorithm then selects the executors who offer the lowest bids, making sure the total cost stays within the budget.**Jen:**  It's like an auction, but with a fixed budget? **Tom:**  You could say that.  And here's the key part – the authors have designed the system to be truthful.**Jen:**  What does that mean?**Tom:**  It means that the task executors have no incentive to lie about their bids.  If they try to inflate their bids to get a higher payment, the algorithm might not choose them, and they'll get nothing.  **Jen:**  So, honesty is the best policy!  That's important for ensuring fairness. **Tom:**  Absolutely!  The authors have done a lot of theoretical analysis to prove that this system is fair and efficient. They've shown that it's computationally efficient, meaning it doesn't take too long to run, and it's individually rational, meaning that the task executors will always make a profit or at least break even.  **Jen:**  This all sounds fantastic! But I'm wondering, how do the researchers know this actually works in the real world?**Tom:**  Good question! They've conducted simulations using real-world data from participatory democracy initiatives.  **Jen:**  So, they're not just theorizing!  **Tom:**  Right!  The simulation results support their conclusions. They found that, as the probability of a project getting funded increases, the number of projects that actually get funded also increases.  **Jen:**  That makes sense.  The more likely people are to vote for a project, the more likely it is to get the funds.**Tom:**  And they also tested the incentive mechanism by comparing their system to other existing mechanisms.  They found that their system consistently outperformed the others in terms of maximizing the overall profit of the task executors, while still staying within the budget constraints.**Reflections****Jen:**  This paper really brings together a lot of interesting elements – democracy, crowdsourcing, and even a bit of game theory to ensure fairness. **Tom:**  I agree.  It's a great example of how technology can be used to empower citizens and create more efficient systems. **Jen:**  I'm also impressed by the authors' thorough analysis. They didn't just throw ideas out there; they actually proved that their system works.**Tom:**  Definitely.  And the fact that they used real-world data adds even more weight to their findings.  **Jen:**  However, I do have a question.  The paper focuses on time-constrained tasks.  What about crowdsensing projects that are not limited by time?  Would this system still work? **Tom:**  That's a great point, Jen.  This paper lays the groundwork, but there's definitely room for future research to explore how this system could be adapted to different types of crowdsensing projects. **Jen:**  And what about the potential for people to manipulate the voting process?  **Tom:**  Another great point!  The authors acknowledge that this is a possible concern and suggest further research to explore ways to mitigate it.  **Conclusion****Tom:**  So, to wrap things up, we've learned about a clever system for incentivizing participation in crowdsensing projects, using a combination of participatory budgeting and a truthful incentive mechanism.  **Jen:**  This research shows the potential for merging technology and democracy to create more efficient and equitable systems.  It's inspiring to see how we can leverage the collective wisdom of the crowd to solve real-world problems.**Tom:**  And it's just the beginning.  As we move forward, we can expect to see even more innovative approaches to crowdsensing, fueled by the principles of transparency, fairness, and citizen empowerment.**Jen:** Thanks for joining us today, and remember, stay curious! We'll see you in the next episode of Research Unraveled.\"\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mtts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:07.535098Z\u001b[0m [\u001b[31m\u001b[1merror    \u001b[0m] \u001b[1mException in tts: 400 Either `input.text` or `input.ssml` is longer than the limit of 5000 bytes. This limit is different from quotas. To fix, reduce the byte length of the characters in this request, or consider using the Long Audio API: https://cloud.google.com/text-to-speech/docs/create-audio-text-long-audio-synthesis.\n",
      "NoneType: None\n",
      "\u001b[0m [\u001b[34m\u001b[1mjonlog\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35merror\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:09.693033Z\u001b[0m [\u001b[31m\u001b[1mcritical \u001b[0m] \u001b[1mGoogleTTS error from text=\"## Episode Title: Crowdsourcing, Democracy, and Dollars: How to Get the Most Out of Your City's Data**Intro****Tom:** Welcome back to Research Unraveled, the podcast where we take complex research papers and break them down for everyone to understand. I'm your host, Tom, and with me, as always, is the amazing Jen. **Jen:** Hey everyone! Today we're diving into a fascinating paper that examines how to make crowdsensing, the practice of gathering data from everyday people, more efficient and fair.  **Tom:**  This isn't just about collecting random data, Jen.  We're talking about harnessing the power of the crowd to tackle real-world problems, like monitoring environmental changes or traffic congestion, and they do it using a system that combines participatory budgeting with incentives. **Jen:** Intriguing! It sounds like this paper is merging technology and democracy in a really unique way.  **Tom:**  It definitely is. The title of the paper is A Participatory Budgeting based Truthful Budget-Limited Incentive Mechanism for Time-Constrained Tasks in Crowdsensing Systems. Quite a mouthful, right? But don't worry, we'll break it all down. **Body****Jen:**  Okay, let's start at the beginning. What's crowdsensing, and why is it so important?**Tom:**  Imagine you want to gather data about traffic patterns in your city, but you don't have the resources to install sensors everywhere.  Crowdsensing offers a solution.  It leverages the fact that people carry smartphones and other devices that can be used as sensors. They can be the eyes and ears of your city, contributing valuable data.**Jen:**  That makes sense! But how do you convince people to participate?  That's where the incentive mechanism comes in, right?**Tom:**  Exactly. The authors of this paper propose a system where the government, or any funding organization, provides a limited budget to support these crowdsensing projects.  This budget is allocated based on a participatory budgeting approach.**Jen:**  What's participatory budgeting?  **Tom:**  Instead of the government deciding which projects get funded, the local citizens get to vote. They express their preferences for projects they think are most important. Think about it like a city-wide poll, where people can weigh in on whether they want more streetlights, better parks, or improved public transportation.**Jen:**  Wow, so it's like a direct democracy for crowdsensing!  That's empowering.  But how do you ensure that the budget is used effectively and that the people contributing data are fairly compensated?**Tom:** This is where things get interesting. The authors have devised a two-tiered system.  **Jen:**  Two tiers?**Tom:**  Yes. First, the participatory budgeting process determines which crowdsensing projects get funding. This involves the city dwellers voting on their preferred projects, and then the government allocates its limited funds based on the votes. **Jen:**  So, it's like a popularity contest where the projects with the most votes get funded?**Tom:**  Not exactly. They have an algorithm to optimize the budget allocation. The algorithm aims to ensure that the projects with the most potential benefit to the city, as perceived by the citizens, get prioritized.  It's a more sophisticated system that takes into account the estimated costs of each project and the total available budget.**Jen:**  Okay, that makes sense.  But what about the second tier?**Tom:**  The second tier is about actually executing the crowdsensing projects.  Once a project is funded, it needs to be divided into tasks that can be performed by the crowd, and those tasks need to be scheduled in a way that doesn't create conflicts.  **Jen:**  Like, you wouldn't want two people trying to collect traffic data at the same spot at the same time, right?**Tom:**  Exactly. They use a scheduling algorithm to make sure that the tasks are assigned to different time slots, preventing overlap and ensuring that all the data is collected efficiently. **Jen:**  So far, it seems like a very well-thought-out system! But I'm curious, how are the people participating in the data collection actually rewarded?**Tom:**  That's where the budget-limited incentive mechanism comes in. Each crowdsensing task has a set budget assigned to it. The people participating, who we call task executors, submit bids for how much they're willing to do the task for. The algorithm then selects the executors who offer the lowest bids, making sure the total cost stays within the budget.**Jen:**  It's like an auction, but with a fixed budget? **Tom:**  You could say that.  And here's the key part – the authors have designed the system to be truthful.**Jen:**  What does that mean?**Tom:**  It means that the task executors have no incentive to lie about their bids.  If they try to inflate their bids to get a higher payment, the algorithm might not choose them, and they'll get nothing.  **Jen:**  So, honesty is the best policy!  That's important for ensuring fairness. **Tom:**  Absolutely!  The authors have done a lot of theoretical analysis to prove that this system is fair and efficient. They've shown that it's computationally efficient, meaning it doesn't take too long to run, and it's individually rational, meaning that the task executors will always make a profit or at least break even.  **Jen:**  This all sounds fantastic! But I'm wondering, how do the researchers know this actually works in the real world?**Tom:**  Good question! They've conducted simulations using real-world data from participatory democracy initiatives.  **Jen:**  So, they're not just theorizing!  **Tom:**  Right!  The simulation results support their conclusions. They found that, as the probability of a project getting funded increases, the number of projects that actually get funded also increases.  **Jen:**  That makes sense.  The more likely people are to vote for a project, the more likely it is to get the funds.**Tom:**  And they also tested the incentive mechanism by comparing their system to other existing mechanisms.  They found that their system consistently outperformed the others in terms of maximizing the overall profit of the task executors, while still staying within the budget constraints.**Reflections****Jen:**  This paper really brings together a lot of interesting elements – democracy, crowdsourcing, and even a bit of game theory to ensure fairness. **Tom:**  I agree.  It's a great example of how technology can be used to empower citizens and create more efficient systems. **Jen:**  I'm also impressed by the authors' thorough analysis. They didn't just throw ideas out there; they actually proved that their system works.**Tom:**  Definitely.  And the fact that they used real-world data adds even more weight to their findings.  **Jen:**  However, I do have a question.  The paper focuses on time-constrained tasks.  What about crowdsensing projects that are not limited by time?  Would this system still work? **Tom:**  That's a great point, Jen.  This paper lays the groundwork, but there's definitely room for future research to explore how this system could be adapted to different types of crowdsensing projects. **Jen:**  And what about the potential for people to manipulate the voting process?  **Tom:**  Another great point!  The authors acknowledge that this is a possible concern and suggest further research to explore ways to mitigate it.  **Conclusion****Tom:**  So, to wrap things up, we've learned about a clever system for incentivizing participation in crowdsensing projects, using a combination of participatory budgeting and a truthful incentive mechanism.  **Jen:**  This research shows the potential for merging technology and democracy to create more efficient and equitable systems.  It's inspiring to see how we can leverage the collective wisdom of the crowd to solve real-world problems.**Tom:**  And it's just the beginning.  As we move forward, we can expect to see even more innovative approaches to crowdsensing, fueled by the principles of transparency, fairness, and citizen empowerment.**Jen:** Thanks for joining us today, and remember, stay curious! We'll see you in the next episode of Research Unraveled.\"\u001b[0m [\u001b[34m\u001b[1m__main__\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35mtts\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n",
      "\u001b[2m2024-05-19T15:58:09.696082Z\u001b[0m [\u001b[31m\u001b[1merror    \u001b[0m] \u001b[1mException in tts: 400 Either `input.text` or `input.ssml` is longer than the limit of 5000 bytes. This limit is different from quotas. To fix, reduce the byte length of the characters in this request, or consider using the Long Audio API: https://cloud.google.com/text-to-speech/docs/create-audio-text-long-audio-synthesis.\n",
      "NoneType: None\n",
      "\u001b[0m [\u001b[34m\u001b[1mjonlog\u001b[0m] \u001b[36mcalling_func\u001b[0m=\u001b[35merror\u001b[0m \u001b[36mthread_id\u001b[0m=\u001b[35m123145723940864\u001b[0m \u001b[36mthread_name\u001b[0m=\u001b[35mThreadPoolExecutor-6_0\u001b[0m\n"
     ]
    }
   ],
   "source": [
    "# %%time\n",
    "sub = 'cs.GT'\n",
    "ep = run(sub, upload=True, limit=5)\n",
    "IPython.display.Audio(merge_mp3s(ep.sounds))\n",
    "\n",
    "# # # # d = '/Users/jong/Documents/PodPapers/Conciousness'\n",
    "# # # # ep = episode_with_pdfs(d)\n",
    "# # # # IPython.display.Audio(merge_mp3s(ep.sounds))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "1dd53bf1-33ce-419c-806b-69cf1c07a369",
   "metadata": {},
   "outputs": [],
   "source": [
    "1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "5c4d6388-9f4a-475e-9a7b-0dfc69251d4b",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3 (ipykernel)",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}