Definition
Plain language
A framing where a broadly capable model is fine-tuned into a specialist without losing its general competence.
As stated in the literature
A training philosophy that takes a generally post-trained foundation model and applies staged specialization toward a target domain while preserving cross-domain capability.
Why it matters: It captures the idea that specialization shouldn't require throwing away general capability, which matters for teams that want one model that's both broad and deeply expert.
For example, you take an instruction-tuned chat model and fine-tune it into a medical-coding specialist that still answers general questions competently.
Heard on the show
“The "specializable-generalist" framing.”Episode 048 — How a 30B Open Model Reached Olympiad Gold With the Right Recipe