AI PM Experience Probe: Prompt Engineering vs. Prompt Tuning vs. Fine-Tuning
Company: Internet Brands
Role: Product Manager
Category: Machine Learning
Difficulty: medium
Interview Round: Other
In a PM interview, you are asked whether you have done prompt fine-tuning. Answer as you would in the room: make clear what you take the term to mean, and describe the work and your part in it.
```hint Untangle the term
The phrase can point to more than one technique. Decide which ones the interviewer might mean before you claim experience with any of them.
```
### Clarifying Questions
- Does the interviewer mean iterating on the prompt text, learning soft prompt vectors while the model stays frozen, or updating the model's weights on task examples?
- Is the interest in work you did hands-on, or in how you scoped, evaluated and decided on that work as the PM?
- Should the answer address a hosted model reached through an API, where weights usually cannot be changed, or a model the team could train itself?
### What a Strong Answer Covers
- A precise distinction between prompt engineering, prompt tuning and fine-tuning, and the problems each one suits.
- An honest, specific account of the candidate's own work, including where the PM's role ended and engineering's began.
- How quality was measured: an evaluation set, success criteria and before-and-after results.
- The cost, latency, data and maintenance trade-offs behind choosing one technique over another.
### Follow-up Questions
- How would you decide that prompt changes have stopped paying off and fine-tuning is worth the cost?
- How do you stop a prompt change that fixes one case from breaking others already in production?
- If you did fine-tune, what data did you train on, and how did you check its quality and handle user privacy?
Overview: An interviewer asks a product manager candidate whether they have done prompt fine-tuning on an LLM product. It tests separating prompt engineering, learned soft prompts and weight fine-tuning, describing one's own work honestly, and explaining the evaluation, cost and data trade-offs behind each choice.