apps/web-crawl-q-and-a/web-qa.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "https://openai.com/\n",
      "https://openai.com/blog/tags/announcements\n",
      "https://openai.com/blog/introducing-openai\n",
      "https://openai.com/blog/authors/ilya\n",
      "https://openai.com/blog/requests-for-research-2\n",
      "https://openai.com/blog/authors/diederik\n",
      "https://openai.com/blog/block-sparse-gpu-kernels\n",
      "https://openai.com/blog/authors/alec\n",
      "https://openai.com/blog/fine-tuning-gpt-2\n",
      "https://openai.com/blog/authors/paul\n",
      "https://openai.com/blog/concrete-ai-safety-problems\n",
      "https://openai.com/blog/learning-to-summarize-with-human-feedback\n",
      "https://openai.com/blog/authors/long\n",
      "https://openai.com/blog/authors/lowe\n",
      "https://openai.com/blog/learning-to-cooperate-compete-and-communicate\n",
      "https://openai.com/blog/authors/jean\n",
      "https://openai.com/blog/authors/igor\n",
      "https://openai.com/blog/neural-mmo\n",
      "https://openai.com/blog/authors/phillip\n",
      "https://openai.com/blog/evolved-policy-gradients\n",
      "https://openai.com/blog/authors/richard\n",
      "https://openai.com/blog/better-exploration-with-parameter-noise\n",
      "https://openai.com/blog/authors/xi\n",
      "https://openai.com/blog/authors/matthias\n",
      "https://openai.com/blog/solving-rubiks-cube\n",
      "https://openai.com/blog/authors/ilge\n",
      "https://openai.com/blog/vpt\n",
      "https://openai.com/blog/authors/brandon\n",
      "https://openai.com/blog/authors/raul\n",
      "https://openai.com/blog/authors/bowen\n",
      "https://openai.com/blog/authors/jie\n",
      "https://openai.com/blog/tags/five\n",
      "https://openai.com/blog/openai-five-benchmark-results\n",
      "https://openai.com/blog/openai-five/#rapid\n",
      "https://openai.com/blog/authors/henrique\n",
      "https://openai.com/blog/authors/susan\n",
      "https://openai.com/blog/authors/brooke\n",
      "https://openai.com/blog/authors/michael-petrov\n",
      "https://openai.com/blog/multimodal-neurons\n",
      "https://openai.com/blog/authors/shan\n",
      "https://openai.com/blog/authors/daniela\n",
      "https://openai.com/blog/authors/nick\n",
      "https://openai.com/blog/authors/chris\n",
      "https://openai.com/blog/introducing-activation-atlases\n",
      "https://openai.com/blog/authors/ludwig-schubert\n",
      "https://openai.com/blog/authors/justin\n",
      "https://openai.com/blog/authors/gabriel\n",
      "https://openai.com/blog/microscope\n",
      "https://openai.com/blog/authors/przemyslaw\n",
      "https://openai.com/blog/authors/david\n",
      "https://openai.com/blog/authors/jakub-pachocki\n",
      "https://openai.com/blog/authors/christy\n",
      "https://openai.com/blog/improving-language-model-behavior\n",
      "https://openai.com/blog/authors/irene\n",
      "https://openai.com/blog/gpt-2-1-5b-release\n",
      "https://openai.com/blog/authors/jack-clark\n",
      "https://openai.com/blog/cooperation-on-safety\n",
      "https://openai.com/blog/authors/amanda\n",
      "https://openai.com/blog/ai-safety-needs-social-scientists\n",
      "https://openai.com/blog/adversarial-example-research\n",
      "https://openai.com/blog/authors/sandy\n",
      "https://openai.com/blog/authors/ian\n",
      "https://openai.com/blog/machine-learning-unconference\n",
      "https://openai.com/events/code-of-conduct.txt\n",
      "https://openai.com/blog/authors/rocky\n",
      "https://openai.com/blog/authors/nicolas\n",
      "https://openai.com/blog/preparing-for-malicious-uses-of-ai\n",
      "https://openai.com/blog/authors/michael\n",
      "https://openai.com/blog/spam-detection-in-the-physical-world\n",
      "https://openai.com/blog/authors/rachel\n",
      "https://openai.com/blog/authors/alex-ray\n",
      "https://openai.com/blog/generalizing-from-simulation\n",
      "https://openai.com/blog/authors/lerrel\n",
      "https://openai.com/blog/authors/xue\n",
      "https://openai.com/blog/faster-robot-simulation-in-python\n",
      "https://openai.com/blog/safety-gym\n",
      "https://openai.com/blog/authors/joshua\n",
      "https://openai.com/blog/spinning-up-in-deep-rl\n",
      "https://openai.com/blog/spinning-up-in-deep-rl-workshop-review\n",
      "https://openai.com/blog/hackathon-follow-up\n",
      "https://openai.com/blog/authors/parnian\n",
      "https://openai.com/blog/openai-hackathon\n",
      "https://openai.com/events/hackathon.txt\n",
      "https://openai.com/blog/authors/josh-tobin\n",
      "https://openai.com/blog/report-from-the-self-organizing-conference\n",
      "https://openai.com/blog/faulty-reward-functions\n",
      "https://openai.com/blog/authors/miles\n",
      "https://openai.com/blog/language-model-safety-and-misuse\n",
      "https://openai.com/blog/authors/tyna\n",
      "https://openai.com/blog/webgpt\n",
      "https://openai.com/blog/authors/jacob-hilton\n",
      "https://openai.com/blog/measuring-goodharts-law\n",
      "https://openai.com/careers/research-engineer\n",
      "https://openai.com/blog/authors/leo\n",
      "https://openai.com/blog/learning-to-summarize-with-human-feedback/#optimizingtherewardmodel\n",
      "https://openai.com/blog/procgen-benchmark\n",
      "https://openai.com/blog/first-retro-contest-retrospective\n",
      "https://openai.com/blog/authors/oleg\n",
      "https://openai.com/blog/roboschool\n",
      "https://openai.com/blog/gym-retro\n",
      "https://openai.com/blog/authors/vicki\n",
      "https://openai.com/blog/retro-contest\n",
      "https://openai.com/blog/authors/alex\n",
      "https://openai.com/blog/reptile\n",
      "https://openai.com/blog/dall-e-2-pre-training-mitigations\n",
      "https://openai.com/blog/authors/larissa\n",
      "https://openai.com/blog/openai-scholars-2018-final-projects\n",
      "https://openai.com/blog/authors/karl\n",
      "https://openai.com/blog/grade-school-math\n",
      "https://openai.com/blog/authors/vineet\n",
      "https://openai.com/blog/authors/christopher\n",
      "https://openai.com/blog/quantifying-generalization-in-reinforcement-learning\n",
      "https://openai.com/blog/authors/reiichiro\n",
      "https://openai.com/blog/authors/suchir\n",
      "https://openai.com/blog/authors/katie\n",
      "https://openai.com/blog/authors/sandhini\n",
      "https://openai.com/blog/authors/pamela\n",
      "https://openai.com/blog/authors/steven\n",
      "https://openai.com/blog/authors/gretchen\n",
      "https://openai.com/blog/authors/jan\n",
      "https://openai.com/blog/critiques\n",
      "https://openai.com/blog/authors/william-saunders\n",
      "https://openai.com/blog/authors/catherine\n",
      "https://openai.com/blog/our-approach-to-alignment-research\n",
      "https://openai.com/blog/best-practices-for-deploying-language-models\n",
      "https://openai.com/blog/instruction-following/#limitations\n",
      "https://openai.com/blog/economic-impacts\n",
      "https://openai.com/blog/authors/sam-manning\n",
      "https://openai.com/scholars\n",
      "https://openai.com/blog/openai-scholars-2021-final-projects\n",
      "https://openai.com/blog/openai-scholars-2020-final-projects\n",
      "https://openai.com/resources\n",
      "https://openai.com/blog/openai-scholars-spring-2020\n",
      "https://openai.com/blog/openai-scholars-class-of-19\n",
      "https://openai.com/blog/openai-scholars-2019-final-projects\n",
      "https://openai.com/blog/authors/jonathan\n",
      "https://openai.com/blog/discovering-types-for-entity-disambiguation\n",
      "https://openai.com/blog/openai-five-benchmark\n",
      "https://openai.com/blog/openai-five-defeats-dota-2-world-champions/#arena\n",
      "https://openai.com/blog/openai-five/#ourapproach\n",
      "https://openai.com/blog/more-on-dota-2/#botexploits\n",
      "https://openai.com/blog/openai-five-benchmark-results/#training\n",
      "https://openai.com/blog/openai-five-finals\n",
      "https://openai.com/five/#overview\n",
      "https://openai.com/blog/dota-2\n",
      "https://openai.com/the-international\n",
      "https://openai.com/blog/more-on-dota-2\n",
      "https://openai.com/blog/the-international-2018-results\n",
      "https://openai.com/blog/openai-five-defeats-dota-2-world-champions/#cooperativemode\n",
      "https://openai.com/blog/openai-five-defeats-dota-2-world-champions\n",
      "https://openai.com/blog/authors/jeff\n",
      "https://openai.com/blog/authors/adrien\n",
      "https://openai.com/blog/authors/joost\n",
      "https://openai.com/blog/authors/peter-zhokhov\n",
      "https://openai.com/blog/authors/glenn\n",
      "https://openai.com/blog/authors/peter\n",
      "https://openai.com/blog/authors/raphael\n",
      "https://openai.com/blog/authors/lilian\n",
      "https://openai.com/blog/techniques-for-training-large-neural-networks\n",
      "https://openai.com/blog/authors/alex-paino\n",
      "https://openai.com/blog/authors/nikolas\n",
      "https://openai.com/blog/openai-five\n",
      "https://openai.com/blog/authors/bob\n",
      "https://openai.com/blog/authors/qiming\n",
      "https://openai.com/blog/authors/wojciech\n",
      "https://openai.com/blog/authors/arthur\n",
      "https://openai.com/blog/authors/mateusz\n",
      "https://openai.com/blog/authors/maciek\n",
      "https://openai.com/blog/authors/jerry\n",
      "https://openai.com/blog/authors/lei\n",
      "https://openai.com/blog/how-to-train-your-openai-five\n",
      "https://openai.com/blog/authors/jonas-schneider\n",
      "https://openai.com/jobs/#robotics\n",
      "https://openai.com/interview-guide\n",
      "https://openai.com/blog/learning-dexterity\n",
      "https://openai.com/blog/authors/rafal\n",
      "https://openai.com/blog/ingredients-for-robotics-research\n",
      "https://openai.com/blog/authors/vikash\n",
      "https://openai.com/blog/authors/marcin\n",
      "https://openai.com/blog/authors/prafulla\n",
      "https://openai.com/blog/authors/szymon-sidor\n",
      "https://openai.com/blog/openai-baselines-dqn\n",
      "https://openai.com/blog/authors/tamim\n",
      "https://openai.com/blog/learning-montezumas-revenge-from-a-single-demonstration\n",
      "https://openai.com/blog/authors/bradly\n",
      "https://openai.com/blog/authors/rein\n",
      "https://openai.com/blog/authors/jonathan-ho\n",
      "https://openai.com/blog/learning-a-hierarchy\n",
      "https://openai.com/blog/authors/peter-chen\n",
      "https://openai.com/blog/authors/kevin\n",
      "https://openai.com/blog/authors/filip\n",
      "https://openai.com/five\n",
      "https://openai.com/blog/authors/yilun\n",
      "https://openai.com/blog/authors/joseph\n",
      "https://openai.com/blog/interpretable-machine-learning-through-teaching\n",
      "https://openai.com/blog/authors/smitha\n",
      "https://openai.com/blog/learning-to-model-other-minds\n",
      "https://openai.com/blog/authors/shimon\n",
      "https://openai.com/blog/authors/maruan\n",
      "https://openai.com/blog/authors/jakob-foerster\n",
      "https://openai.com/blog/nonlinear-computation-in-linear-networks\n",
      "https://openai.com/blog/energy-based-models\n",
      "https://openai.com/blog/emergent-tool-use\n",
      "https://openai.com/blog/authors/ingmar\n",
      "https://openai.com/blog/authors/todor\n",
      "https://openai.com/blog/learning-concepts-with-energy-functions\n",
      "https://openai.com/blog/authors/yi\n",
      "https://openai.com/blog/authors/pieter\n",
      "https://openai.com/blog/authors/aviv\n",
      "https://openai.com/blog/instruction-following\n",
      "https://openai.com/blog/learning-to-communicate\n",
      "https://openai.com/blog/authors/jon\n",
      "https://openai.com/blog/summarizing-books\n",
      "https://openai.com/blog/authors/chelsea\n",
      "https://openai.com/blog/gathering_human_feedback\n",
      "https://openai.com/blog/authors/dario-amodei\n",
      "https://openai.com/blog/science-of-ai\n",
      "https://openai.com/blog/authors/jared\n",
      "https://openai.com/blog/authors/sam\n",
      "https://openai.com/blog/gpt-2-6-month-follow-up\n",
      "https://openai.com/blog/better-language-models/#update\n",
      "https://openai.com/blog/authors/david-luan\n",
      "https://openai.com/blog/authors/danny\n",
      "https://openai.com/blog/ai-and-efficiency\n",
      "https://openai.com/blog/authors/david-lansky\n",
      "https://openai.com/blog/authors/tom\n",
      "https://openai.com/blog/testing-robustness\n",
      "https://openai.com/blog/authors/jacob\n",
      "https://openai.com/blog/authors/yi-sun\n",
      "https://openai.com/blog/authors/daniel\n",
      "https://openai.com/blog/authors/dan\n",
      "https://openai.com/blog/deep-reinforcement-learning-from-human-preferences\n",
      "https://openai.com/blog/authors/geoffrey\n",
      "https://openai.com/blog/debate\n",
      "https://openai.com/blog/authors/jeffrey\n",
      "https://openai.com/blog/authors/nisan\n",
      "https://openai.com/blog/amplifying-ai-training\n",
      "https://openai.com/blog/authors/daniel-ziegler\n",
      "https://openai.com/blog/baselines-acktr-a2c\n",
      "https://openai.com/blog/authors/yuhuai\n",
      "https://openai.com/blog/authors/shun\n",
      "https://openai.com/blog/authors/elman\n",
      "https://openai.com/blog/openai-baselines-ppo\n",
      "https://openai.com/blog/language-unsupervised\n",
      "https://openai.com/blog/tags/baselines\n",
      "https://openai.com/blog/authors/scott\n",
      "https://openai.com/blog/sparse-transformer\n",
      "https://openai.com/blog/authors/rewon\n",
      "https://openai.com/blog/glow\n",
      "https://openai.com/blog/authors/john\n",
      "https://openai.com/blog/openai-gym-beta\n",
      "https://openai.com/blog/authors/tim\n",
      "https://openai.com/jobs\n",
      "https://openai.com/blog/formal-math\n",
      "https://openai.com/blog/authors/stanislas\n",
      "https://openai.com/blog/authors/jesse\n",
      "https://openai.com/blog/generative-models\n",
      "https://openai.com/blog/authors/andrej\n",
      "https://openai.com/blog/distill\n",
      "https://openai.com/blog/authors/vicki-cheung\n",
      "https://openai.com/blog/jukebox\n",
      "https://openai.com/projects/five\n",
      "https://openai.com/blog/authors/christine\n",
      "https://openai.com/blog/authors/jong\n",
      "https://openai.com/blog/authors/heewoo\n",
      "https://openai.com/blog/musenet\n",
      "https://openai.com/blog/better-language-models\n",
      "https://openai.com/blog/robots-that-learn\n",
      "https://openai.com/blog/authors/ankur\n",
      "https://openai.com/blog/authors/erika-reinhardt\n",
      "https://openai.com/blog/deep-double-descent\n",
      "https://openai.com/blog/authors/tristan\n",
      "https://openai.com/blog/authors/preetum\n",
      "https://openai.com/blog/authors/boaz\n",
      "https://openai.com/blog/authors/yamini\n",
      "https://openai.com/blog/authors/gal\n",
      "https://openai.com/blog/tags/gpt-2\n",
      "https://openai.com/blog/clip\n",
      "https://openai.com/blog/ai-and-compute\n",
      "https://openai.com/blog/authors/girish\n",
      "https://openai.com/blog/special-projects\n",
      "https://openai.com/blog/authors/sam-altman\n",
      "https://openai.com/blog/unsupervised-sentiment-neuron\n",
      "https://openai.com/blog/dall-e\n",
      "https://openai.com/blog/authors/aditya\n",
      "https://openai.com/blog/authors/mark\n",
      "https://openai.com/blog/authors/mikhail\n",
      "https://openai.com/blog/authors/vedant\n",
      "https://openai.com/blog/competitive-self-play\n",
      "https://openai.com/blog/authors/trapit\n",
      "https://openai.com/blog/meta-learning-for-wrestling\n",
      "https://openai.com/blog/authors/yura\n",
      "https://openai.com/blog/reinforcement-learning-with-prediction-based-rewards\n",
      "https://openai.com/blog/authors/harri\n",
      "https://openai.com/blog/image-gpt\n",
      "https://openai.com/blog/evolution-strategies\n",
      "https://openai.com/blog/infrastructure-for-deep-learning\n",
      "https://openai.com/blog/generative-models/#gan\n",
      "https://openai.com/blog/generative-models#improving-gans\n",
      "https://openai.com/blog/tags/multimodal\n",
      "https://openai.com/gpt-3\n",
      "https://openai.com/javascript:setMathjaxCookie()\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/abs/2005.14165v1\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/list/cs.CL/new\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/abs/2005.14165v3\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/auth/show-endorsers/2005.14165\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/list/cs/recent\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/abs/2005.14165?context=cs\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/{url_path('ignore_me')}\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/abs/2005.14165v2\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/show-email/b5cb66e9/2005.14165\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/prevnext?id=2005.14165&function=next&context=cs.CL\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/format/2005.14165\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/prevnext?id=2005.14165&function=prev&context=cs.CL\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/pdf/2005.14165\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/tb/2005.14165\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/list/cs.CL/2005\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/list/cs.CL/recent\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/blog/dall-e-2\n",
      "https://openai.com/blog/authors/openai\n",
      "https://openai.com/blog/improving-verifiability\n",
      "https://openai.com/blog/dall-e-2-extending-creativity\n",
      "https://openai.com/blog/the-international\n",
      "https://openai.com/blog/symposium-2019\n",
      "https://openai.com/blog/tags/culture\n",
      "https://openai.com/blog/learning-day\n",
      "https://openai.com/blog/openai-fellows-fall-2018\n",
      "https://openai.com/blog/neurips-2020\n",
      "https://openai.com/blog/tags/community\n",
      "https://openai.com/blog/universe\n",
      "https://openai.com/blog/openai-gym-beta/#rl\n",
      "https://openai.com/blog/openai-technical-goals/#goal4\n",
      "https://openai.com/blog/authors/elon\n",
      "https://openai.com/blog/scaling-kubernetes-to-7500-nodes\n",
      "https://openai.com/blog/scaling-kubernetes-to-2500-nodes\n",
      "https://openai.com/blog/authors/christopher-berner\n",
      "https://openai.com/blog/authors/bchess\n",
      "https://openai.com/blog/authors/eric\n",
      "https://openai.com/blog/forecasting-misuse\n",
      "https://openai.com/forecasting-misuse-paper\n",
      "https://openai.com/prevnext?id=2301.04246&function=prev&context=cs.CY\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/auth/show-endorsers/2301.04246\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/format/2301.04246\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/pdf/2301.04246\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/show-email/64c5c6bd/2301.04246\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/list/cs.CY/recent\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/prevnext?id=2301.04246&function=next&context=cs.CY\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/list/cs.CY/new\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/list/cs.CY/2301\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/abs/2301.04246?context=cs\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/blog/authors/greg\n",
      "https://openai.com/blog/dall-e-api-now-available-in-public-beta\n",
      "https://openai.com/blog/api-no-waitlist\n",
      "https://openai.com/blog/dall-e-introducing-outpainting\n",
      "https://openai.com/blog/team-update\n",
      "https://openai.com/blog/chatgpt-plus\n",
      "https://openai.com/blog/openai-api\n",
      "https://openai.com/jobs/#applied-ai\n",
      "https://openai.com/blog/authors/mira\n",
      "https://openai.com/join\n",
      "Unable to parse page https://openai.com/join due to JavaScript being required\n",
      "HTTP Error 403: Forbidden\n",
      "https://openai.com/blog/tags/residency\n",
      "https://openai.com/blog/openai-licenses-gpt-3-technology-to-microsoft\n",
      "https://openai.com/blog/microsoft\n",
      "https://openai.com/blog/team-update-august\n",
      "https://openai.com/blog/new-ai-classifier-for-indicating-ai-written-text\n",
      "https://openai.com/blog/authors/lama\n",
      "https://openai.com/blog/authors/scott-aaronson\n",
      "https://openai.com/blog/authors/jan-hendrik-kirchner\n",
      "https://openai.com/blog/tags/api\n",
      "https://openai.com/blog/openai-fellows\n",
      "https://openai.com/blog/tags/scholars\n",
      "https://openai.com/blog/openai-and-microsoft-extend-partnership\n",
      "https://openai.com/blog/dall-e-now-available-without-waitlist\n",
      "https://openai.com/blog/helen-toner-joins\n",
      "https://openai.com/blog/team-update-january\n",
      "https://openai.com/blog/team-plus-plus#interns\n",
      "https://openai.com/blog/openai-codex\n",
      "https://openai.com/blog/openai-scholars-2019\n",
      "https://openai.com/blog/authors/ashley\n",
      "https://openai.com/blog/openai-scholars\n",
      "https://openai.com/blog/dall-e-now-available-in-beta\n",
      "https://openai.com/blog/new-and-improved-embedding-model\n",
      "https://openai.com/blog/authors/ryan\n",
      "https://openai.com/blog/authors/arvind\n",
      "https://openai.com/blog/authors/ted\n",
      "https://openai.com/blog/dall-e-2-update\n",
      "https://openai.com/blog/authors/joanne\n",
      "https://openai.com/blog/tags/fellows\n",
      "https://openai.com/blog/openai-summer-fellows-2018\n",
      "https://openai.com/blog/authors/maddie\n",
      "https://openai.com/blog/codex-apps\n",
      "https://openai.com/blog/codex\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/blog/new-and-improved-content-moderation-tooling\n",
      "https://openai.com/blog/authors/teddy\n",
      "https://openai.com/blog/authors/angela\n",
      "https://openai.com/blog/authors/chong\n",
      "https://openai.com/blog/welcome-pieter-and-shivon\n",
      "https://openai.com/blog/openai-technical-goals\n",
      "https://openai.com/blog/procgen-minerl-competitions\n",
      "https://openai.com/blog/will-hurd-joins\n",
      "https://openai.com/blog/fund\n",
      "https://openai.com/news\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/news/introducing-our-first-investments\n",
      "HTTP Error 404: Not Found\n",
      "https://openai.com/blog/introducing-text-and-code-embeddings\n",
      "https://openai.com/blog/authors/boris\n",
      "https://openai.com/blog/openai-scholars-2018-meet-our-scholars\n",
      "https://openai.com/blog/team-plus-plus\n",
      "https://openai.com/blog/gpt-3-apps\n",
      "https://openai.com/jobs/#open\n",
      "https://openai.com/blog/customized-gpt-3\n",
      "https://openai.com/blog/authors/luke\n",
      "https://openai.com/blog/authors/rachel-lim\n",
      "https://openai.com/blog/authors/michael-wu\n",
      "https://openai.com/blog/openai-supporters\n",
      "https://openai.com/blog/openai-residency\n",
      "https://openai.com/blog/leadership-team-update\n",
      "https://openai.com/blog/organizational-update\n",
      "https://openai.com/blog/openai-fellows-interns-2019\n",
      "https://openai.com/blog/openai-scholars-2020\n",
      "https://openai.com/blog/gpt-3-edit-insert\n",
      "https://openai.com/blog/authors/mo\n",
      "https://openai.com/blog/openai-pytorch\n",
      "https://openai.com/blog/openai-scholars-2019-meet-our-scholars\n",
      "https://openai.com/blog/openai-charter\n",
      "https://openai.com/blog/openai-and-microsoft\n",
      "https://openai.com/blog/openai-lp\n",
      "https://openai.com/blog/reducing-bias-and-improving-safety-in-dall-e-2\n",
      "https://openai.com/terms\n",
      "https://openai.com/api/policies/service-terms\n",
      "https://openai.com/api/policies/sharing-publication\n",
      "https://openai.com/api/policies/terms\n",
      "https://openai.com/security/disclosure\n",
      "https://openai.com/blog/whisper\n",
      "https://openai.com/blog/authors/tao\n",
      "https://openai.com/research\n",
      "https://openai.com/api/docs\n",
      "Unable to parse page https://openai.com/api/docs due to JavaScript being required\n",
      "HTTP Error 403: Forbidden\n",
      "https://openai.com/dall-e-2\n",
      "https://openai.com/privacy\n",
      "https://openai.com/api\n",
      "https://openai.com/blog\n",
      "https://openai.com/blog/triton\n",
      "https://openai.com/blog/authors/philippe\n",
      "https://openai.com/jobs/#acceleration\n",
      "https://openai.com/blog/robust-adversarial-inputs\n",
      "https://openai.com/blog/authors/anish-athalye\n",
      "https://openai.com/blog/tags/milestones\n",
      "https://openai.com/alignment\n",
      "https://openai.com\n",
      "https://openai.com/publications\n",
      "https://openai.com/charter\n",
      "https://openai.com/blog/tags/research\n",
      "https://openai.com/fund\n",
      "https://openai.com/about\n",
      "https://openai.com/timeline\n",
      "https://openai.com/careers\n",
      "https://openai.com/api/examples\n",
      "Unable to parse page https://openai.com/api/examples due to JavaScript being required\n",
      "HTTP Error 403: Forbidden\n",
      "https://openai.com/api/login\n",
      "Unable to parse page https://openai.com/api/login due to JavaScript being required\n",
      "HTTP Error 403: Forbidden\n",
      "https://openai.com/newsroom\n",
      "https://openai.com/api/policies\n",
      "https://openai.com/api/pricing\n",
      "https://openai.com/contact-sales\n",
      "https://openai.com/api/pricing/#faq-fine-tuning-pricing-calculation\n",
      "https://openai.com/blog/tags/events\n",
      "https://openai.com/blog/chatgpt\n"
     ]
    }
   ],
   "source": [
    "import requests\n",
    "import re\n",
    "import urllib.request\n",
    "from bs4 import BeautifulSoup\n",
    "from collections import deque\n",
    "from html.parser import HTMLParser\n",
    "from urllib.parse import urlparse\n",
    "import os\n",
    "\n",
    "# Regex pattern to match a URL\n",
    "HTTP_URL_PATTERN = r'^http[s]*://.+'\n",
    "\n",
    "# Define root domain to crawl\n",
    "domain = \"openai.com\"\n",
    "full_url = \"https://openai.com/\"\n",
    "\n",
    "# Create a class to parse the HTML and get the hyperlinks\n",
    "class HyperlinkParser(HTMLParser):\n",
    "    def __init__(self):\n",
    "        super().__init__()\n",
    "        # Create a list to store the hyperlinks\n",
    "        self.hyperlinks = []\n",
    "\n",
    "    # Override the HTMLParser's handle_starttag method to get the hyperlinks\n",
    "    def handle_starttag(self, tag, attrs):\n",
    "        attrs = dict(attrs)\n",
    "\n",
    "        # If the tag is an anchor tag and it has an href attribute, add the href attribute to the list of hyperlinks\n",
    "        if tag == \"a\" and \"href\" in attrs:\n",
    "            self.hyperlinks.append(attrs[\"href\"])\n",
    "\n",
    "# Function to get the hyperlinks from a URL\n",
    "def get_hyperlinks(url):\n",
    "    \n",
    "    # Try to open the URL and read the HTML\n",
    "    try:\n",
    "        # Open the URL and read the HTML\n",
    "        with urllib.request.urlopen(url) as response:\n",
    "\n",
    "            # If the response is not HTML, return an empty list\n",
    "            if not response.info().get('Content-Type').startswith(\"text/html\"):\n",
    "                return []\n",
    "            \n",
    "            # Decode the HTML\n",
    "            html = response.read().decode('utf-8')\n",
    "    except Exception as e:\n",
    "        print(e)\n",
    "        return []\n",
    "\n",
    "    # Create the HTML Parser and then Parse the HTML to get hyperlinks\n",
    "    parser = HyperlinkParser()\n",
    "    parser.feed(html)\n",
    "\n",
    "    return parser.hyperlinks\n",
    "\n",
    "# Function to get the hyperlinks from a URL that are within the same domain\n",
    "def get_domain_hyperlinks(local_domain, url):\n",
    "    clean_links = []\n",
    "    for link in set(get_hyperlinks(url)):\n",
    "        clean_link = None\n",
    "\n",
    "        # If the link is a URL, check if it is within the same domain\n",
    "        if re.search(HTTP_URL_PATTERN, link):\n",
    "            # Parse the URL and check if the domain is the same\n",
    "            url_obj = urlparse(link)\n",
    "            if url_obj.netloc == local_domain:\n",
    "                clean_link = link\n",
    "\n",
    "        # If the link is not a URL, check if it is a relative link\n",
    "        else:\n",
    "            if link.startswith(\"/\"):\n",
    "                link = link[1:]\n",
    "            elif link.startswith(\"#\") or link.startswith(\"mailto:\"):\n",
    "                continue\n",
    "            clean_link = \"https://\" + local_domain + \"/\" + link\n",
    "\n",
    "        if clean_link is not None:\n",
    "            if clean_link.endswith(\"/\"):\n",
    "                clean_link = clean_link[:-1]\n",
    "            clean_links.append(clean_link)\n",
    "\n",
    "    # Return the list of hyperlinks that are within the same domain\n",
    "    return list(set(clean_links))\n",
    "\n",
    "\n",
    "def crawl(url):\n",
    "    # Parse the URL and get the domain\n",
    "    local_domain = urlparse(url).netloc\n",
    "\n",
    "    # Create a queue to store the URLs to crawl\n",
    "    queue = deque([url])\n",
    "\n",
    "    # Create a set to store the URLs that have already been seen (no duplicates)\n",
    "    seen = set([url])\n",
    "\n",
    "    # Create a directory to store the text files\n",
    "    if not os.path.exists(\"text/\"):\n",
    "            os.mkdir(\"text/\")\n",
    "\n",
    "    if not os.path.exists(\"text/\"+local_domain+\"/\"):\n",
    "            os.mkdir(\"text/\" + local_domain + \"/\")\n",
    "\n",
    "    # Create a directory to store the csv files\n",
    "    if not os.path.exists(\"processed\"):\n",
    "            os.mkdir(\"processed\")\n",
    "\n",
    "    # While the queue is not empty, continue crawling\n",
    "    while queue:\n",
    "\n",
    "        # Get the next URL from the queue\n",
    "        url = queue.pop()\n",
    "        print(url) # for debugging and to see the progress\n",
    "\n",
    "        # Save text from the url to a <url>.txt file\n",
    "        with open('text/'+local_domain+'/'+url[8:].replace(\"/\", \"_\") + \".txt\", \"w\") as f:\n",
    "\n",
    "            # Get the text from the URL using BeautifulSoup\n",
    "            soup = BeautifulSoup(requests.get(url).text, \"html.parser\")\n",
    "\n",
    "            # Get the text but remove the tags\n",
    "            text = soup.get_text()\n",
    "\n",
    "            # If the crawler gets to a page that requires JavaScript, it will stop the crawl\n",
    "            if (\"You need to enable JavaScript to run this app.\" in text):\n",
    "                print(\"Unable to parse page \" + url + \" due to JavaScript being required\")\n",
    "            \n",
    "            # Otherwise, write the text to the file in the text directory\n",
    "            f.write(text)\n",
    "\n",
    "        # Get the hyperlinks from the URL and add them to the queue\n",
    "        for link in get_domain_hyperlinks(local_domain, url):\n",
    "            if link not in seen:\n",
    "                queue.append(link)\n",
    "                seen.add(link)\n",
    "\n",
    "crawl(full_url)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "def remove_newlines(serie):\n",
    "    serie = serie.str.replace('\\n', ' ')\n",
    "    serie = serie.str.replace('\\\\n', ' ')\n",
    "    serie = serie.str.replace('  ', ' ')\n",
    "    serie = serie.str.replace('  ', ' ')\n",
    "    return serie"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "\n",
    "# Create a list to store the text files\n",
    "texts=[]\n",
    "\n",
    "# Get all the text files in the text directory\n",
    "for file in os.listdir(\"text/\" + domain + \"/\"):\n",
    "\n",
    "    # Open the file and read the text\n",
    "    with open(\"text/\" + domain + \"/\" + file, \"r\") as f:\n",
    "        text = f.read()\n",
    "\n",
    "        # Omit the first 11 lines and the last 4 lines, then replace -, _, and #update with spaces.\n",
    "        texts.append((file[11:-4].replace('-',' ').replace('_', ' ').replace('#update',''), text))\n",
    "\n",
    "# Create a dataframe from the list of texts\n",
    "df = pd.DataFrame(texts, columns = ['fname', 'text'])\n",
    "\n",
    "# Set the text column to be the raw text with the newlines removed\n",
    "df['text'] = df.fname + \". \" + remove_newlines(df.text)\n",
    "df.to_csv('processed/scraped.csv')\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot: >"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import tiktoken\n",
    "\n",
    "# Load the cl100k_base tokenizer which is designed to work with the ada-002 model\n",
    "tokenizer = tiktoken.get_encoding(\"cl100k_base\")\n",
    "\n",
    "df = pd.read_csv('processed/scraped.csv', index_col=0)\n",
    "df.columns = ['title', 'text']\n",
    "\n",
    "# Tokenize the text and save the number of tokens to a new column\n",
    "df['n_tokens'] = df.text.apply(lambda x: len(tokenizer.encode(x)))\n",
    "\n",
    "# Visualize the distribution of the number of tokens per row using a histogram\n",
    "df.n_tokens.hist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "max_tokens = 500\n",
    "\n",
    "# Function to split the text into chunks of a maximum number of tokens\n",
    "def split_into_many(text, max_tokens = max_tokens):\n",
    "\n",
    "    # Split the text into sentences\n",
    "    sentences = text.split('. ')\n",
    "\n",
    "    # Get the number of tokens for each sentence\n",
    "    n_tokens = [len(tokenizer.encode(\" \" + sentence)) for sentence in sentences]\n",
    "    \n",
    "    chunks = []\n",
    "    tokens_so_far = 0\n",
    "    chunk = []\n",
    "\n",
    "    # Loop through the sentences and tokens joined together in a tuple\n",
    "    for sentence, token in zip(sentences, n_tokens):\n",
    "\n",
    "        # If the number of tokens so far plus the number of tokens in the current sentence is greater \n",
    "        # than the max number of tokens, then add the chunk to the list of chunks and reset\n",
    "        # the chunk and tokens so far\n",
    "        if tokens_so_far + token > max_tokens:\n",
    "            chunks.append(\". \".join(chunk) + \".\")\n",
    "            chunk = []\n",
    "            tokens_so_far = 0\n",
    "\n",
    "        # If the number of tokens in the current sentence is greater than the max number of \n",
    "        # tokens, go to the next sentence\n",
    "        if token > max_tokens:\n",
    "            continue\n",
    "\n",
    "        # Otherwise, add the sentence to the chunk and add the number of tokens to the total\n",
    "        chunk.append(sentence)\n",
    "        tokens_so_far += token + 1\n",
    "\n",
    "    return chunks\n",
    "    \n",
    "\n",
    "shortened = []\n",
    "\n",
    "# Loop through the dataframe\n",
    "for row in df.iterrows():\n",
    "\n",
    "    # If the text is None, go to the next row\n",
    "    if row[1]['text'] is None:\n",
    "        continue\n",
    "\n",
    "    # If the number of tokens is greater than the max number of tokens, split the text into chunks\n",
    "    if row[1]['n_tokens'] > max_tokens:\n",
    "        shortened += split_into_many(row[1]['text'])\n",
    "    \n",
    "    # Otherwise, add the text to the list of shortened texts\n",
    "    else:\n",
    "        shortened.append( row[1]['text'] )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot: >"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "df = pd.DataFrame(shortened, columns = ['text'])\n",
    "df['n_tokens'] = df.text.apply(lambda x: len(tokenizer.encode(x)))\n",
    "df.n_tokens.hist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>n_tokens</th>\n",
       "      <th>embeddings</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>blog authors maddie.   Maddie Hall - OpenAI   ...</td>\n",
       "      <td>175</td>\n",
       "      <td>[-0.012958061881363392, -0.006103983614593744,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>blog authors tom.   Tom Brown - OpenAI        ...</td>\n",
       "      <td>228</td>\n",
       "      <td>[-0.0053874170407652855, -0.009962032549083233...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>blog openai scholars 2019 final projects.   Op...</td>\n",
       "      <td>492</td>\n",
       "      <td>[0.0019150723237544298, -0.0070442273281514645...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>In this project, I used curiosity-driven explo...</td>\n",
       "      <td>478</td>\n",
       "      <td>[-0.0067560747265815735, 0.0004431474662851542...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Results revealed that the optimal RL policies ...</td>\n",
       "      <td>499</td>\n",
       "      <td>[-0.012868616729974747, 0.0029640409629791975,...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                text  n_tokens  \\\n",
       "0  blog authors maddie.   Maddie Hall - OpenAI   ...       175   \n",
       "1  blog authors tom.   Tom Brown - OpenAI        ...       228   \n",
       "2  blog openai scholars 2019 final projects.   Op...       492   \n",
       "3  In this project, I used curiosity-driven explo...       478   \n",
       "4  Results revealed that the optimal RL policies ...       499   \n",
       "\n",
       "                                          embeddings  \n",
       "0  [-0.012958061881363392, -0.006103983614593744,...  \n",
       "1  [-0.0053874170407652855, -0.009962032549083233...  \n",
       "2  [0.0019150723237544298, -0.0070442273281514645...  \n",
       "3  [-0.0067560747265815735, 0.0004431474662851542...  \n",
       "4  [-0.012868616729974747, 0.0029640409629791975,...  "
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import openai\n",
    "\n",
    "df['embeddings'] = df.text.apply(lambda x: openai.Embedding.create(input=x, engine='text-embedding-ada-002')['data'][0]['embedding'])\n",
    "df.to_csv('processed/embeddings.csv')\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>n_tokens</th>\n",
       "      <th>embeddings</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>blog authors maddie.   Maddie Hall - OpenAI   ...</td>\n",
       "      <td>175</td>\n",
       "      <td>[-0.012958061881363392, -0.006103983614593744,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>blog authors tom.   Tom Brown - OpenAI        ...</td>\n",
       "      <td>228</td>\n",
       "      <td>[-0.0053874170407652855, -0.009962032549083233...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>blog openai scholars 2019 final projects.   Op...</td>\n",
       "      <td>492</td>\n",
       "      <td>[0.0019150723237544298, -0.0070442273281514645...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>In this project, I used curiosity-driven explo...</td>\n",
       "      <td>478</td>\n",
       "      <td>[-0.0067560747265815735, 0.0004431474662851542...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Results revealed that the optimal RL policies ...</td>\n",
       "      <td>499</td>\n",
       "      <td>[-0.012868616729974747, 0.0029640409629791975,...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                text  n_tokens  \\\n",
       "0  blog authors maddie.   Maddie Hall - OpenAI   ...       175   \n",
       "1  blog authors tom.   Tom Brown - OpenAI        ...       228   \n",
       "2  blog openai scholars 2019 final projects.   Op...       492   \n",
       "3  In this project, I used curiosity-driven explo...       478   \n",
       "4  Results revealed that the optimal RL policies ...       499   \n",
       "\n",
       "                                          embeddings  \n",
       "0  [-0.012958061881363392, -0.006103983614593744,...  \n",
       "1  [-0.0053874170407652855, -0.009962032549083233...  \n",
       "2  [0.0019150723237544298, -0.0070442273281514645...  \n",
       "3  [-0.0067560747265815735, 0.0004431474662851542...  \n",
       "4  [-0.012868616729974747, 0.0029640409629791975,...  "
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from openai.embeddings_utils import distances_from_embeddings\n",
    "\n",
    "df['embeddings'] = df.text.apply(lambda x: openai.Embedding.create(input=x, engine='text-embedding-ada-002')['data'][0]['embedding'])\n",
    "\n",
    "df.to_csv('processed/embeddings.csv')\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>text</th>\n",
       "      <th>n_tokens</th>\n",
       "      <th>embeddings</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>blog authors maddie.   Maddie Hall - OpenAI   ...</td>\n",
       "      <td>175</td>\n",
       "      <td>[-0.012958061881363392, -0.006103983614593744,...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>blog authors tom.   Tom Brown - OpenAI        ...</td>\n",
       "      <td>228</td>\n",
       "      <td>[-0.0053874170407652855, -0.009962032549083233...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>blog openai scholars 2019 final projects.   Op...</td>\n",
       "      <td>492</td>\n",
       "      <td>[0.0019150723237544298, -0.0070442273281514645...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>In this project, I used curiosity-driven explo...</td>\n",
       "      <td>478</td>\n",
       "      <td>[-0.0067560747265815735, 0.0004431474662851542...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>Results revealed that the optimal RL policies ...</td>\n",
       "      <td>499</td>\n",
       "      <td>[-0.012868616729974747, 0.0029640409629791975,...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                                text  n_tokens  \\\n",
       "0  blog authors maddie.   Maddie Hall - OpenAI   ...       175   \n",
       "1  blog authors tom.   Tom Brown - OpenAI        ...       228   \n",
       "2  blog openai scholars 2019 final projects.   Op...       492   \n",
       "3  In this project, I used curiosity-driven explo...       478   \n",
       "4  Results revealed that the optimal RL policies ...       499   \n",
       "\n",
       "                                          embeddings  \n",
       "0  [-0.012958061881363392, -0.006103983614593744,...  \n",
       "1  [-0.0053874170407652855, -0.009962032549083233...  \n",
       "2  [0.0019150723237544298, -0.0070442273281514645...  \n",
       "3  [-0.0067560747265815735, 0.0004431474662851542...  \n",
       "4  [-0.012868616729974747, 0.0029640409629791975,...  "
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "from openai.embeddings_utils import distances_from_embeddings, cosine_similarity\n",
    "\n",
    "df=pd.read_csv('processed/embeddings.csv', index_col=0)\n",
    "df['embeddings'] = df['embeddings'].apply(eval).apply(np.array)\n",
    "\n",
    "df.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'No, you are not allowed to publish model outputs to Twitter without a human review. You must manually review each generation before sharing or while streaming, and indicate that the content is AI-generated in a way no user could reasonably miss or misunderstand.'"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def create_context(\n",
    "    question, df, max_len=1800, size=\"ada\"\n",
    "):\n",
    "    \"\"\"\n",
    "    Create a context for a question by finding the most similar context from the dataframe\n",
    "    \"\"\"\n",
    "\n",
    "    # Get the embeddings for the question\n",
    "    q_embeddings = openai.Embedding.create(input=question, engine='text-embedding-ada-002')['data'][0]['embedding']\n",
    "\n",
    "    # Get the distances from the embeddings\n",
    "    df['distances'] = distances_from_embeddings(q_embeddings, df['embeddings'].values, distance_metric='cosine')\n",
    "\n",
    "\n",
    "    returns = []\n",
    "    cur_len = 0\n",
    "\n",
    "    # Sort by distance and add the text to the context until the context is too long\n",
    "    for i, row in df.sort_values('distances', ascending=True).iterrows():\n",
    "        \n",
    "        # Add the length of the text to the current length\n",
    "        cur_len += row['n_tokens'] + 4\n",
    "        \n",
    "        # If the context is too long, break\n",
    "        if cur_len > max_len:\n",
    "            break\n",
    "        \n",
    "        # Else add it to the text that is being returned\n",
    "        returns.append(row[\"text\"])\n",
    "\n",
    "    # Return the context\n",
    "    return \"\\n\\n###\\n\\n\".join(returns)\n",
    "\n",
    "def answer_question(\n",
    "    df,\n",
    "    model=\"text-davinci-003\",\n",
    "    question=\"Am I allowed to publish model outputs to Twitter, without a human review?\",\n",
    "    max_len=1800,\n",
    "    size=\"ada\",\n",
    "    debug=False,\n",
    "    max_tokens=150,\n",
    "    stop_sequence=None\n",
    "):\n",
    "    \"\"\"\n",
    "    Answer a question based on the most similar context from the dataframe texts\n",
    "    \"\"\"\n",
    "    context = create_context(\n",
    "        question,\n",
    "        df,\n",
    "        max_len=max_len,\n",
    "        size=size,\n",
    "    )\n",
    "    # If debug, print the raw model response\n",
    "    if debug:\n",
    "        print(\"Context:\\n\" + context)\n",
    "        print(\"\\n\\n\")\n",
    "\n",
    "    try:\n",
    "        # Create a completions using the question and context\n",
    "        response = openai.Completion.create(\n",
    "            prompt=f\"Answer the question based on the context below, and if the question can't be answered based on the context, say \\\"I don't know\\\"\\n\\nContext: {context}\\n\\n---\\n\\nQuestion: {question}\\nAnswer:\",\n",
    "            temperature=0,\n",
    "            max_tokens=max_tokens,\n",
    "            top_p=1,\n",
    "            frequency_penalty=0,\n",
    "            presence_penalty=0,\n",
    "            stop=stop_sequence,\n",
    "            model=model,\n",
    "        )\n",
    "        return response[\"choices\"][0][\"text\"].strip()\n",
    "    except Exception as e:\n",
    "        print(e)\n",
    "        return \"\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "\"I don't know.\""
      ]
     },
     "execution_count": 33,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "answer_question(df, question=\"What day is it?\", debug=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'The newest embeddings model is text-embedding-ada-002.'"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "answer_question(df, question=\"What is our newest embeddings model?\")"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "env",
   "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": "3.9.6"
  },
  "orig_nbformat": 4,
  "vscode": {
   "interpreter": {
    "hash": "05f34a34d73b71652304030c1097be3a5720ea2447153dd6542d145a26b73181"
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}