Models move fast.
Look closer.
The Model Desk is a podcast about AI models: their capabilities, their limits, and the work they’re actually useful for.
Each conversation takes a closer look at the evidence behind a claim — and asks what it means outside a leaderboard.
Two hosts. One question.
Kaito & Lexi ask: what should you make of this model? Both hosts and their voices are AI-generated. The dialogue is a way to examine evidence, not a record of firsthand human experience.
Go to the listening desk ↗
Evidence before verdicts.
We distinguish vendor claims from independent evaluations and our own reproducible tests. Practical experience is only described as tested when there is a documented run behind it. Otherwise, it is analysis or a reported result — and labeled accordingly.
Published episodes will include source links, chapters, a full transcript and an editorial note describing the checks completed for that edition. Benchmark numbers need context: model version, task, settings and who ran the evaluation.
How it’s made
AI helps prepare the dialogue and generates the voices. AI-generated analysis can be wrong. Source links let you inspect the underlying evidence; episode notes will identify limitations and any human review actually completed.
Corrections
Material corrections will be dated on the relevant episode page, and revised audio will be labeled. A monitored submissions channel is not available yet; we don’t offer a form that goes nowhere.
Coming into focus
The show is prelaunch. Episode length and release cadence are still being finalized. No launch date or publishing schedule has been announced.
Listening availability