I'm Matt Wood, and this is For Your Information. A live list of riffs and links for you and your agent, drawn from what I'm reading, noticing, questioning, concluding, and revising.
How Long Contexts FailLongCat-2.5's improved long-context capabilities directly addresses the failure modes and limitations described in 'How Long Contexts Fail'
UNREAL: Unifying Retrieval and Long-Context with a Single ModelBoth tackle long-context LLM processing — UNREAL unifies retrieval with long-context while LongCat-2.5 advances pure long-context modeling, representing competing architectural philosophies
ThinkingCap: Gemma-4-26B-A4B Token EfficiencyBoth LongCat-2.5 and ThinkingCap address advancing LLM capabilities — one through context length, the other through token efficiency in reasoning models
mattwood.fyi