TOPIC
4 Stories Published
language models
A.I.
8m
Bolmo and Byteification: What Nature’s New Language-Model Paper Changes About Tokenizers
Nature's October 7, 2026 paper on byteification converts existing subword language models to read raw bytes using under 1% of pretraining compute.…
A.I.
9m
Why Transformers Stop Following References Early, and How a Rank-8 LoRA at One Layer Pushes Them Further
A September 29 arXiv paper reports that pretrained transformers follow only a few in-context references, and that a tiny LoRA at one…
A.I.
9m
Why AI Chatbots Hallucinate, and Can It Be Fixed?
Chatbots invent facts because of how they are trained and graded. The causes, notable cases and the fixes being tried.
Research
8m
Mechanistic Interpretability Under Test: Do Circuit Explanations Tell the Whole Story?
A master's-thesis preprint scores circuit-based explanations in GPT-2 Small. They read well and are sufficient, but cover only about 22% of the…