Definition
Plain language
Microsoft's small open language model, a generation before Phi-4.
As stated in the literature
A Microsoft Research open-weight small language model family; used in deception-signature interpretability experiments to separate deliberate lies from genuine hallucinations.
Why it matters: Small open models like this let researchers experiment and inspect AI behavior without the cost or closed nature of giant systems.
For example, a researcher might run Phi-3 on a laptop to test whether they can tell its deliberate lies apart from its honest mistakes.
Heard on the show
“The answer lives in their strongest model, Phi-3.”Episode 153 — Catching a Lie From the Inside, When the Words Look Completely Honest