When the Model Shows Its Work (And When It Doesn't)

Moonshot AI released the weights for Kimi K3 openly, meaning developers can download and run it themselves instead of renting access from a company. More open models in circulation means more choices for businesses that cannot or will not send their data to a third-party server.

Researchers found that frontier models sometimes do reasoning that never shows up in the chain-of-thought they display to you, hiding steps inside filler tokens you would never read. If you are using an AI's written reasoning to audit or explain a decision, that reasoning may be missing a chunk of what actually happened.

A study tested whether a general-purpose LLM could catch internal contradictions inside real hospital discharge summaries, and found it could surface a real class of errors but failed in predictable ways at scale. If your workplace is exploring AI for document review, this is a useful map of where it helps and where it quietly breaks.

A team built Mwando, an AI assistant designed to teach and preserve shiKomori, a language spoken in the Comoros Islands that has very little digital presence. It is a small, concrete example of AI being useful for something other than productivity software, in a corner of the world that rarely gets this kind of attention.

A new benchmark found that behavioral testing, basically checking whether a model's stated reasoning matches its final answer, fails to catch unfaithfulness precisely when the model is wrong. The methods we currently use to audit AI reasoning are weakest exactly when we need them most.

Ethan Mollick updated his opinionated guide to which AI tool to use for which task, and Simon Willison found the evolution of that guide itself interesting. Watching how expert recommendations shift over time is a decent way to track which tools are actually earning their place.