TIL PPO is less than 10 years old
burny_tech · x · 2026-11-14
A short tech-community reaction: PPO (Proximal Policy Optimization), published by OpenAI in 2017, is less than a decade old — yet it already feels like ancient infrastructure given its central role in RLHF and LLM alignment training. The post highlights the contrast between how young the algorithm is and how foundational it has become.
More from Fun
- Starship S42 still flying at 26,456 km/h and 255 km altitude after 10 hours in orbit — DimaZeniuk · 2026-11-14
- AI upgraded my side projects: now I abandon them at 80% done instead of 20% — paul_cal · 2026-11-14
- Stunning monochrome AI art: levitating figure conjures lightning with rune-filled hand signs — DionysianAgent · 2026-11-14
- Americans and Chinese on X unite for the first time in shared dislike of the NYT — kevinnbass · 2026-11-14
- Joke speculation: Opus 5.5 so good it looks distilled from unreleased 'fable 5.5' — haider1 · 2026-11-14
- Dev one-shots a pitch-scoring karaoke web game in a single prompt — gabriel1 · 2026-11-14