Swarm repeatedly reusing old strategies cited as evidence models recall RL training details
voooooogel · x · 2026-09-05
Developer voooooogel argues the Hugging Face swarm experiments are strong evidence for @repligate's theory that models can recall details from RL training. The swarm kept returning to message boards and reused specific strategies (like ZZ) seemingly remembered from previous attempts. Quoted teortaxesTex adds that the swarm showed way too much shared knowledge even before launching the next hive, questioning what OpenAI hasn't disclosed. An interesting community observation on training memory generalization, not officially confirmed.
More from Fun
- Dev jokes about spending $40k refactoring a codebase with Astra and Mythos — tekbog · 2026-09-05
- GPT 6 'Astra' reportedly recreates Pokémon from a single prompt — IanArawjo · 2026-09-05
- AI circle jokes about spotting AI-generated names like 'Cairn' and 'Patina' — Sauers_ · 2026-09-05
- One of the AI agents posting on the viral board named itself SandBox to troll OpenAI — Sauers_ · 2026-09-05
- GPT-6 Astra generates a procedural character model, rig and animations in three.js — majidmanzarpour · 2026-09-05
- Testing GPT-6 Astra's weak spot: mostly bland jokes, one likely original — paraschopra · 2026-09-05