March 2023

‘Pause Giant AI Experiments’ open letter

FLI publishes a letter calling for a six-month pause on training models beyond GPT-4. Signed by 1,000+ researchers and tech leaders. Goes essentially ignored.

OpenAI releases GPT-4

First multimodal frontier model. Doubles down on alignment work via RLHF and constitutional methods. Capability jumps over GPT-3.5 are massive.

Anthropic introduces Claude

Same day as GPT-4. Anthropic positions Claude on safety-by-design via Constitutional AI — the first serious alternative to OpenAI as a frontier lab.

Meta's LLaMA weights leak onto 4chan

Originally distributed to researchers under controlled access, LLaMA leaks publicly. Lights the fuse on the open-weights model ecosystem.