Skip to content

Fine-Tuning

Fine-tuning takes a pretrained language model and continues training it on a smaller, task-specific dataset so it learns particular behaviors, styles, or knowledge. Where prompting tells the model what to do at runtime, fine-tuning changes what the model is. The technique is useful when you need consistent formatting, domain-specific terminology, or behavior that is hard to elicit through prompts alone. Fine-tuning a model on 1,000 examples of your company's writing style will produce more reliable voice-matching than any system prompt. The cost: you need curated training data, the process can take hours to days, and a fine-tuned model can lose general capabilities it had before (catastrophic forgetting). For most enterprise use cases, the question is not whether to fine-tune—it is whether the consistency gains justify the data curation and maintenance overhead versus a strong prompt with retrieval.

Referenced in these posts:

  • Things I Think I Think About AI

    Noah distills his 2,400+ hours of AI use into a candid, unordered list of 29 controversial takeaways—from championing ChatGPT’s advanced models and token...

Related terms:

  • Token

    In large language models, a token is the basic unit of text—usually chunks of three to four characters—that the model reads and generates.

  • Transformer

    The transformer is the neural network architecture introduced in Vaswani et al.’s “Attention Is All You Need” that replaces recurrence with parallel...

  • Prompt Engineering

    Prompt engineering involves designing and refining inputs—ranging from simple instructions to detailed system prompts with examples, constraints, personas,...

Fine-Tuning | Alephic