Awesome
this list is now archived - look inside https://github.com/sw-yx/brain if you would like an updated list! thank you.
gpt3-list
List of things that people are claiming is enabled by GPT3. unverified but links to sources.
Official info
- Official paper https://arxiv.org/abs/2005.14165
- GitHub: https://github.com/openai/gpt-3
- HN: https://news.ycombinator.com/item?id=23345379
- criticism https://www.lesswrong.com/posts/ZHrpjDc3CepSeeBuE/gpt-3-a-disappointing-paper
- nvidia summary https://news.developer.nvidia.com/openai-presents-gpt-3-a-175-billion-parameters-language-model/
- venturebeat summary https://venturebeat.com/2020/05/29/openai-debuts-gigantic-gpt-3-language-model-with-175-billion-parameters/
- how it works visualized/animated https://jalammar.github.io/how-gpt3-works-visualizations-animations/
- Vicki Boykis on CommonCrawl and WebText https://vicki.substack.com/p/are-you-as-smart-as-a-toddler
- beta signup https://beta.openai.com/
another list like this one https://gpt-3.is/
Notable takes
- "racecar for the mind" https://twitter.com/zebulgar/status/1284294275430117376?s=20
- polluting gpt4 https://twitter.com/karpathy/status/1284660899198820352?s=20
- precision and specification https://twitter.com/JoeHughesDev/status/1284985137298575366?s=20
- more applications https://nesslabs.com/gpt-3-future-productivity
- makes SWEs more valuable https://davnicwil.com/ai-writing-code-makes-software-engineers-more-valuable/
- "beginning of the end" https://kitze.io/posts/gpt3-is-the-beginning-of-the-end
- Astroturfing risk https://jayriverlong.github.io/2020/07/24/gpt3.html
GPT-3 marks a new generation of tools enabling states to interfere in online discourse with unprecedented scale and persuasiveness, enabling great damage at low cost.
- a16z What's hype, What's real: https://a16z.com/2020/07/29/16mins-gpt3-natural-language-neural-net-deep-learning-few-shot-zero-shot/
- Tempering expectations https://minimaxir.com/2020/07/gpt3-expectations/
However, a 30-40% success rate implies a 60-70% failure rate, which is patently unsuitable for a production application. If it takes seconds to generate a React component and it takes on average 3 tries to get something usable, it might be more pragmatic to just create the component the hard, boring way. Compare again to Apple’s Siri, which can get very frustrating when it performs the wrong action.
GPT3 responding to GPT3 takes: https://twitter.com/raphamilliere/status/1289129723310886912?s=20
<details> <summary>Yann LeCun take endorsing Nabla health's article
</summary> </details>More:
- https://twitter.com/michael_nielsen/status/1284937254666768384?s=20
- https://maraoz.com/2020/07/18/openai-gpt3/
- https://youtube.com/watch?v=SY5PvZrJhLE&feature=youtu.be
- https://youtube.com/watch?v=S0KakHcj_rs
- Andrew Ng https://youtube.com/watch?v=SysgYptB198
- @AnalyticsVidhya: https://analyticsvidhya.com/blog/2019/06/understanding-transformers-nlp-state-of-the-art-models/
- @JayAlammar: http://jalammar.github.io/illustrated-transformer/
Projects/Startups
https://trends.vc/trends-0057-ai-generated-content/
- AI Dungeon https://twitter.com/nickwalton00/status/1283143210999484416?s=20
- codegen
- AI Channels https://andrewmayneblog.wordpress.com/2020/06/11/the-aichannels-project/
- Suhail browser autocomplete https://twitter.com/Suhail/status/1284562825084416000?s=20
- Search Engine https://twitter.com/paraschopra/status/1284801028676653060?s=20
- QuickResponse (Email completion) https://twitter.com/OthersideAI/status/1285776335638614017?s=20
- Lossless (english to regex) https://losslesshq.com/
- Producing Customer Feedback Insight https://askviable.com/ (Rahul Vohra demo)
- Replit Code Oracle - code querying https://twitter.com/amasad/status/1285959600584265729?s=20
- Explain Like I'm Five https://twitter.com/chris__lu/status/1284535970164088832?s=20
- Taglines.ai https://twitter.com/chris__lu/status/1290426666477592576?s=20
- AI email composer https://compose.ai/
- AI video creator https://www.synthesia.io/
- https://www.copysmith.ai/ "This startup is using GPT-3 for content marketing, just tried it out and it's surprisingly good at generating 2-3 sentence pitches (going off of some key words and a brief description)"
- Philosopher.ai philosopher prompting
Code
- HTML/Layout
- Functioning React code
- Small button counters https://twitter.com/sharifshameem/status/1284095222939451393?s=20
- Todo-list https://twitter.com/sharifshameem/status/1284421499915403264?s=20
- based on variable name https://twitter.com/hturan/status/1282261783147958272?s=20
- SQL
- Keras Models
- Swift UI
- shell commands
- regex
Design
Finance
- financial statements https://twitter.com/itsyashdani/status/1285695850300219392?s=20
Music
Image completion
Therapy
Domain specific Text
- Medical questions: https://twitter.com/xuenay/status/1283376153156362242?s=20
- Law: https://twitter.com/f_j_j_/status/1283349995144359937
- Presentations: https://twitter.com/bemmu/status/1285284131656445952?s=20
General Text
- Fortune cookies
- Poetry
- Creative Fiction
- https://www.gwern.net/GPT-3
- NPC text?
- Non English Languages
- Emulating famous people
- Essay completion
- GPT3 vs GPT3
Failures
- Turing test
- http://lacker.io/ai/2020/07/06/giving-gpt-3-a-turing-test.html (origin https://twitter.com/nabeelqu/status/1284167539141087232?s=20)
- you can warn about nonsense qtns https://twitter.com/xuenay/status/1284053165483143172?s=20
- you can give uncertainty prompts "yo be real" https://arr.am/2020/07/25/gpt-3-uncertainty-prompts/
- https://aiweirdness.com/post/621186154843324416/all-your-questions-answered
- inhuman response times https://www.kmeme.com/2020/10/gpt-3-bot-went-undetected-askreddit-for.html?m=1
- http://lacker.io/ai/2020/07/06/giving-gpt-3-a-turing-test.html (origin https://twitter.com/nabeelqu/status/1284167539141087232?s=20)
- Technical Interviews
- memory
- Bias