Fine-tuning changes behaviour, not every limitation
Fine-tuning adjusts how a model responds. It does not repair what the underlying model cannot do.
It is good at tone, format and domain-specific phrasing. It is not a reliable way to install facts, and it does not make the model stop generating plausible inventions. Teams frequently reach for it expecting accuracy and get style. Retrieval, verification and constraining what the system may act on are the tools for the other problem.
More on AI and LLMs
- An LLM predicts plausible continuations, not verified truthIt only checks the shape
- Context is temporary working material, not permanent knowledgeThe board gets wiped
- Retrieval adds documents, not guaranteed correctnessThe filter slot is empty
- Temperature changes variation, not factualityThe dial only sets the spread
- System prompts are instructions, not a security boundaryA sign, with no fence
- LLM output is untrusted input downstreamIt comes in round the back