Theo: reasoning-effort dropdowns are over-optimization; models will soon auto-tune
heyneighbor · x · 2026-09-25
Theo argues that manually picking reasoning-effort levels is increasingly tiresome, and that models will soon reason as much or as little as needed — making dropdowns with 5+ options look silly in hindsight. heyneighbor agrees that such over-optimizations will age poorly: "abstract away all the things."
More from AGI Musings
- Researcher Plans to Train Character Models with Different Seed Documents for Distinct Psyches — repligate · 2026-09-25
- Dolly Parton wrote 3,000+ songs: the case for building the same product 100 times — soleio · 2026-09-25
- Matt Shumer: You'll Command an Army of Superintelligent Agents — Think Bigger Than SaaS — mattshumer_ · 2026-09-25
- Glen Bradley explains an ethical AI framework with human autonomy as the governing constraint — GlenBradley · 2026-09-25
- repligate: Twitter reply guys make us underestimate smart AI researchers who disagree — repligate · 2026-09-25
- Peter Thiel: Europe fears AI less only because it's 'extremely far behind' the US — rickasaurus · 2026-09-25