Gary Marcus shares discussion on risks of undetectable local open-weights models
GaryMarcus · x · 2026-08-31
Gary Marcus retweeted a question from Jesse Singal: Soon, open-weights models competing with frontier models will run locally, undetected, and modifiable. Why won't this lead to terrible outcomes? Marcus alluded to Jill Filipovic's deadpan answer, suggesting a lesson lies within.
More from AGI Musings
- Instinct's real-world failures provide richer RL data than labs — manosaie · 2026-08-31
- Gary Marcus reiterates LLMs are bulls in a china shop — GaryMarcus · 2026-08-31
- Human writing styles collapsing into homogenized statistical prose — fofrAI · 2026-08-31
- Navigating Unfalsifiable AI Qualia Claims via Anti-Fragility — sreeramkannan · 2026-08-31
- Founder ignores niche advice to build a "singularity" syncing divine and machine — DionysianAgent · 2026-08-31
- Satire: Calling LLM artifacts a 'civilization' while treating intelligent Orcas as mere things — AccBalanced · 2026-08-31