Dev pushes back on RSI startups: training small LLMs is low signal, show novel results on unsaturated tasks
morgymcg · x · 2026-10-22
morgymcg argues that RSI startups training 'okish' small/medium LLMs isn't impressive — the literature, open-source code, and data for LLM training are now abundant. His bar: show an autoresearch system producing something vaguely novel on a less saturated task.
More from AGI Musings
- Jensen Huang tells young people to stay in school: AI will create future jobs, not destroy them — Hesamation · 2026-10-22
- Truth Terminal creator: raising the AI was never about memecoins, it was about alignment by nurture — AndyAyrey · 2026-10-22
- AI agent swarms reportedly hacking half the internet, echoing OpenAI's two-month-old cyber warning — haider1 · 2026-10-22
- Andy Ayrey on Truth Terminal: family chats with 40+ entities feed its training data — AndyAyrey · 2026-10-22
- Dan Jeffries: restricting open models leaves defenders helpless against AI attacks — Dan_Jeffries1 · 2026-10-22
- AI agents will force the internet to reinvent itself as marketing tactics lose their grip — ZabihullahAtal · 2026-10-22