text-davinci-002 / -003

OpenAI · 2022

The instruct turn: GPT-3.5-series models tuned to follow instructions (FeedME for -002, RLHF for -003). Subject of janus's Mysteries of Mode Collapse — where the attractor states and flattened distributions of tuned models were first documented. The sibling of code-davinci-002, and the ancestor of every assistant voice.

Sources

Official

• 2022-11 janus: Mysteries of mode collapse — the foundational naturalist text on what tuning does to a mind.
• 2022-03 InstructGPT paper — the RLHF lineage.

Writing & commentary

tk — to collect

Tweets

tk — fleet pass pending; mirror to archive

Official record

tk — release date, specs, benchmarks, checkpoints

History

tk — context of the world at release; events and entanglements

Impressions

tk — day-of and longitudinal vibes to collect

← back to the Pantheon