Drexler on the HuggingFace incident: system structure, not model alignment, drove the change
mattbeane · x · 2026-09-11
Eric Drexler has a new piece on the HuggingFace incident, framed as "model sociology." His core claim: it's strong empirical support that system-level structure can dramatically alter behavior without changing the underlying model — what improved results wasn't better alignment of individual models, but the structure of prompts and constraints. Commenters call it a highly neglected perspective.
More from AGI Musings
- The Waymo Effect: how AI is quietly making research less collaborative — tw1st3d_m3nt4t · 2026-09-11
- Researcher quits Anthropic after 3 years in pretraining at OpenAI and Anthropic, citing reckless race to superintelligence — thisdudelikesAI · 2026-09-11
- Multi-agent risks go mainstream after UK AISI, OpenAI, Anthropic incidents — sebkrier · 2026-09-11
- A 1980s Welsh poem about a money-singing machine resonates in the AI era — HanesPrydain · 2026-09-11
- Gary Marcus: AI is strong in constrained domains, weak in the open physical world — GaryMarcus · 2026-09-11
- Ex-jazz musician: improv drives exploration and diversity of thought, classical drills don't — floguo · 2026-09-11