Thought experiment asks: what are the limits of language modeling when vision itself is language modeling?
rishit_dagli · x · 2026-10-12
Rishit Dagli publishes an article, 'A Thought Experiment on the Limits of Language Modeling,' probing whether current training paradigms impose limits on modeling human language.
- Inspired by LLMs' strength across robotics, video, and 3D/4D generation, he argues that almost all of vision and graphics is actually a language modeling task.
- Modeling is formalized as learning P(X), but 'human language' is loosely defined, requiring care.
- He frames intelligence as a process—innovative data collection, filtering, training recipes, and training code—that could be pointed at anything.
- The piece uses this thought experiment to ask what inherent limits arise from modeling language with today's paradigms.
More from Research
- Radical Numerics details data mixing for DNA+language pretraining of Omnii — aaron_lou · 2026-10-13
- DINOv3 vision transformers learn word meaning from images alone, study finds — udmrzn · 2026-10-13
- Study finds AI models recognize mental health crises but fail to actually help — LinkedInNews · 2026-10-13
- ML reanalysis of historic plates confirms transients cluster within 24h of nuclear tests — coherence · 2026-10-13
- Dev turns EmbeddingGemma 2 into an on-device STT model, transcribing 1 hour of audio in 13 seconds — solyarisoftware · 2026-10-13
- Testing AI agents with simulations: replay real traces, 50 curated sims beat 1000 random ones — HamelHusain · 2026-10-13