Model Bottlenecks Lie in Optimization, Not Expressivity
jasondeanlee · x · 2026-07-18
The core argument of this reply is that the limitations of many current models or recurrent architectures might be bottlenecked by optimization rather than expressive power.
The author's points include:
- "All models are universal," meaning their expressive capacity is largely equivalent.
- Their generalization bounds are also roughly the same.
- Therefore, if approaches like looped transformers underperform, the issue likely stems from insufficient training or suboptimal optimization, rather than the architectural ceiling of expressivity.
This is a concise, research-oriented take, primarily responding to a discussion about a paper on latent reasoning or looped models.
More from Research
- Researchers identify "Forking": train loss falls while validation loss turns up — ZimingLiu11 · 2026-10-26
- Google DeepMind pilots world's first double-blind evaluations for frontier models — scychan_brains · 2026-10-26
- Rewriting scaling law study: optimal synthetic data share rises from 10% to 32% with compute — heghbalz · 2026-10-26
- FLASH: GPU-native deformable sim learns cloth folding in minutes on one RTX 5090 — ChongZzZhang · 2026-10-26
- LEMON: zero-shot LLM misbehavior monitor flags hallucinations from attention maps — yoavartzi · 2026-10-26
- Behind the EMNLP keynote: LMs as simulated learners to generate child language acquisition hypotheses — najoungkim · 2026-10-26