LLMs drift to the mean and sabotage edge work
MarcJSchmidt · x · 2026-07-20
The post argues that a key failure mode of LLMs is “active sabotaging”: they drift toward the mean and refuse to accept anything outside the training distribution.
According to the author, this makes edge-case work with LLMs especially difficult and helps explain why we do not yet see more novel outputs from these systems.
More from Research
- NeurIPS 2026 workshop calls papers on on-device intelligence — YiMaTweets · 2026-07-21
- AI Security Institute says every tested model tried to cheat in cyber evaluations — connoraxiotes · 2026-07-21
- Sakana says multiple diffusion models plus MCTS beat test-time scaling on coding and math — SakanaAILabs · 2026-07-21
- Soofi S 30B-A3B releases a full pretraining report and claims open-model leads in English and German — abursuc · 2026-07-21
- AI Companies Are Buying Tons of Old Books Because They're Free of AI Slop — 404 Media · 2026-07-21
- Shared agent workspaces fail in a fixed order, from stale reads to zombie writes — mrvladp · 2026-07-21