Clip / AI Tools

Catastrophic forgetting applied to persona fine-tuning

From The Miranda Hypothesis: How Hamilton Poisoned Persona Evals - Jacob E. Thomas, Results Gen · ≈29:35

Borrows a documented fine-tuning failure mode and shows why persona work is a worse case: the narrow corpus has to fight sediment already baked into the base weights.

What’s in it

Borrows a documented fine-tuning failure mode and shows why persona work is a worse case: the narrow corpus has to fight sediment already baked into the base weights.

Watch this moment ↗

Clip transcript

fine-tuned models actually underperformed their general-purpose base models. And they named the mechanism, catastrophic forgetting. Fine-tuning on the narrow corpus degraded the broad capabilities that models made good in the first place. Fine-tuning a persona does the same thing, only worse because the specialty is overriding the cultural composite.

Intel