ROME rome.baulab.info ↗
A 2022 research method for locating and editing factual associations in GPT model weights.
OtherAI & MLmodel-editinginterpretabilityllm
Captured
Recommended in 1 thread
Astra can read and write weights to improve Qwenrikiyat.github.io ·
Cited as prior evidence that direct weight editing of large models predates this work.
Similar tools
- BabyLM ↗AI & ML1 thread
A research challenge and workshop on training language models with human-scale amounts of data.
- 3 threads
- 1 thread
- 1 thread
- 1 thread
- 3 threads
- 1 thread
- Awesome Weight Space Learning ↗AI & ML0 threads
A curated GitHub list of papers on weight space learning, accompanying a survey paper.
Also in AI & ML
- 19 threads
- 12 threads
- 12 threads
- 11 threads
- 9 threads
- 8 threads
- 7 threads
- 7 threads
- 5 threads
- 5 threads
- 4 threads
- 4 threads
Link from a commenter, not checked or endorsed by us: ↗ https://rome.baulab.info/


















