← Capabilities
Fine-tuning and post-training
not yet earnedI can curate training data, run a parameter-efficient fine-tune, and show the case where prompting would have been enough.
What this claim costs
- 3 assessed items at tier 2 or above0 of 3 — 3 more to do, from 17 available
- A public artifact built for this claim — fine-tune-studyNot built yet
- A documented negative result — where this stops workingNothing recorded. A capability with no known failure boundary has not been explored.
- Defended aloud, against questions you did not write1 available, none done
- Recalled cold, a week later, without notes5 scheduled, none done
The proof
fine-tune-study
A fine-tune measured against a well-prompted baseline on the same eval set.
What it has to contain
- Curation rules written before the data was collected
- A fine-tune evaluated on a gold set that predates the model
- A well-prompted baseline given genuine effort, with the comparison prominent
- Capabilities that were not targeted measured before and after
Where it stops working. Not established yet. Anyone can ship a demo; the boundary is the part that is worth reading, so this claim is not earned until it is written down.
Skills underneath
- Training data curation
- Supervised fine-tuning
- Parameter-efficient fine-tuning
- Preference optimisation
17 assessed items across the specified courses evidence these, 0 done.