Why LLMs prefer compressed language: the information density hypothesis
code_star · x · 2026-10-23
@thomashikaru proposes a hypothesis: LLMs face pressure for information density, leading to compressed language, novel n-grams, and a preference for short alternatives in linguistic alternations — unlike humans, whose pressure is successful communication with other humans. @codestar responds that the better question is why: is this a beneficial product of performance squeezing or an unintended artifact of some training stage.
More from Research
- EMNLP paper: diffusion LMs internally encode latent denoising timestep representations — gsarti_ · 2026-10-23
- KU Leuven and VIB open two Group Leader professorships in AI for Biology, start 2027 — jpillowtime · 2026-10-23
- Subspace Tuning curbs LLM catastrophic forgetting by freezing principal singular directions — scottniekum · 2026-10-23
- Perceptic launches PercepticOS, an AI workspace for 'scientific superintelligence' — nathanbenaich · 2026-10-23
- EPFL's Bosselut to talk long context as test-time learning at ELLIS seminar — andre_t_martins · 2026-10-23
- DRATS: Adaptive Task Sampling Boosts Worst-Task Performance in Multi-Task RL — Cohere · 2026-10-23