Facts · 2022
events: 14 · Timeline
- Anthropic describes Constitutional AI: a harmless assistant is trained without human labels of harmful outputs, and the only human oversight is a list of rules and principlesScience · Large language model · arxiv.org ↗
- ChatGPT launchesEducation · Work & jobs · Media & information · Chatbots · openai.com ↗
- Programmers file a US class action against GitHub, Microsoft and OpenAI over Copilot — the first case challenging the training and output of AI systemsSoftware development · Laws & regulation · AI pair programmer · githubcopilotlitigation.com ↗
- The US Commerce Department's BIS restricts sales to China of advanced AI chips and semiconductor manufacturing equipmentGeopolitics & defense · Laws & regulation · Compute export controls · bis.gov ↗
- Princeton and Google researchers publish ReAct: a language model interleaves reasoning with actions and revises its plan based on the resultsSoftware development · Science · Autonomous AI agents · arxiv.org ↗
- Cloudflare releases Turnstile, a free CAPTCHA replacement without puzzles; by then the company had cut the number of its own CAPTCHAs shown by 91%Cybersecurity · Privacy & surveillance · Internet & infrastructure · Death of CAPTCHA · blog.cloudflare.com ↗
- Whisper: open speech recognitionMedia & information · Software development · Speech recognition · en.wikipedia.org ↗
- The US government informs NVIDIA that exports of the A100 and H100 to China, Hong Kong and Russia now require a licenseGeopolitics & defense · Business & markets · Compute export controls · sec.gov ↗
- Stable Diffusion is released publiclyCulture & arts · Software development · Generative AI · actuia.com ↗
- Felix Krause shows that TikTok's in-app browser for iOS injects code into third-party websites that subscribes to every keystroke; TikTok replies that the code is used only for debuggingPrivacy & surveillance · Cybersecurity · Behavioral biometrics · krausefx.com ↗
- Midjourney opens its betaCulture & arts · Generative AI · en.wikipedia.org ↗
- MIT Technology Review: Worldcoin's first 450,000 users in 24 countries were lured with cash, AirPods and the promise of tokens, without being told it was a testPrivacy & surveillance · Work & jobs · Proof of personhood · technologyreview.com ↗
- DeepMind presents Chinchilla: with the same compute budget, a 70-billion-parameter model trained on more text outperforms the 280-billion-parameter Gopher; model size and the number of training tokens should be scaled equallyScience · Large language model · arxiv.org ↗
- OpenAI describes InstructGPT: the model is fine-tuned on human ratings (RLHF), and people prefer the answers of its 1.3-billion-parameter version to those of the 175-billion-parameter GPT-3Science · Learning from human feedback (RLHF) · arxiv.org ↗