Writing

Fine-tuning isn't dead, it's just not the default anymore

In-context learning won the spotlight from today's decoder LLMs, but fine-tuning still wins on consistency, control and deep domain knowledge.

Fine-tuning isn’t dead. It’s just not the default anymore.

Back in the BERT days, fine-tuning was the way to adapt a language model for a specific task. But with today’s powerful decoder-based LLMs, in-context learning approaches (like direct prompting, RAG, and whatever else you can think of) have taken the spotlight, and for good reason: they work really well.

That said, fine-tuning still has its place. Yes, you can often get by with prompts, but when you need consistency, control, or deep domain knowledge, fine-tuning might still be the better choice.

Here are a few cases where fine-tuning shines:

  • Custom chatbots – Want your bot to speak in your brand’s voice or tone? Fine-tuning helps lock that in.
  • Text classifiers – Prompts can guide outputs, but fine-tuning (especially on a strong base model) gives better reliability.
  • Domain-specific translation – Legal, medical, or technical jargon? Prompting helps, but fine-tuning really understands the lingo.
  • Low-resource languages – Fine-tuning can boost performance when there’s not much data or support for the language.

So no, fine-tuning isn’t dead. It’s just hanging out in the toolbox, waiting for the right job.