TMS Paper Is Under-Appreciated: Polysemanticity Lives in Weak Activations
livgorton · x · 2026-10-17
The author argues that much of the Toy Models of Superposition paper remains under-appreciated despite its core ideas being widely known, with many of its predictions only now being discussed. She highlights one insight that matches what LM interpretability researchers routinely find: MLP neurons appear interpretable at their top activations, while polysemanticity lives in the weaker ones.
More from Research
- LLMs pick 7 in 30/30 tries: EMNLP paper traces the bias to SFT training data — radamihalcea · 2026-10-17
- OpenAI researcher: hashing an adversary-chosen timestamp doesn't stop rigging — csvoss · 2026-10-17
- Is SWA the frontier secret sauce? DeepSeek and Kimi chose the hard independent route — burny_tech · 2026-10-17
- Evolving competing neural cellular automata show lifelike self-organizing behavior — burny_tech · 2026-10-17
- Vals AI moves recursive self-improvement forecast to July 2027 as Opus 5.5 tops RSI index — 141_1337 · 2026-10-17
- HF Transformers adds activation checkpoint offloading to CPU to free GPU memory — StasBekman · 2026-10-17