Wakaru open-sources ontologize: labeling datasets from model activations
amplifiedamp · x · 2026-10-27
amplifiedamp shares experiments built on WakaruLabs' open-source repo ontologize, using Opus 5.5, an RTX 3090, and Trinity 400B TrueBase activation API to have Claude ontologize a dataset from model activations.
On this dataset, the Wakaru ontologizer underperforms traditional methods on FVU and reconstruction, but produces significantly more useful labels by spreading differentiating capacity across heads. The MIT-licensed repo includes full experiment scripts for SAE training, autointerp, and evals.
More from Research
- New paper: LLM data schedules and learning rate schedules should be co-designed — jasondeanlee · 2026-10-27
- What makes research get adopted: important problems, proper evals, and usability — srchvrs · 2026-10-27
- Bitter lesson part 2: few academic inventions matter, and conference publications aren't the key — srchvrs · 2026-10-27
- First-week PhD student lands ARR Area Chair role, exposing peer-review quirks — anmarasovic · 2026-10-27
- EMNLP keynote: industry does read academic NLP papers, it just rarely credits them — srchvrs · 2026-10-27
- UK AISI: under $80 of fine-tuning fully restores filtered bio capabilities in open-weight models — connoraxiotes · 2026-10-27