csvoss concedes steering vectors are a valid superset: every feature is steerable, not every steerable thing is a feature
csvoss · x · 2026-10-13
csvoss clarified that she originally meant only SAE features, which are clearly mechinterp — but she's happy to concede that steering vectors are a valid superset: every feature is steerable, but not every steerable thing is a feature.
The post quotes her earlier reply that steering vectors are useful but not yet productized, like GPS before ubiquitous smartphones.
Related event: OpenAI researcher: steering vectors useful but not yet productized(3 posts)→
More from Research
- 'Some Simple Economics of AGI' by Catalini, Wei & Wu hailed as the best recent paper on AGI economics — sudoraohacker · 2026-10-13
- BenchCAD 2.0: best frontier model scores just 0.59/1.0 on professional CAD modeling — hanjie_chen · 2026-10-13
- Planning humanoid falls: IROS 2015 paper treats falling as dynamic programming over contacts — chris_j_paxton · 2026-10-13
- BlackboxNLP 2026 program goes live: EMNLP workshop in Budapest on Oct 29 — gsarti_ · 2026-10-13
- Label-free minibatch OT coupling beats conditional OT, hits 1.91 FID on ImageNet-256 — AlexanderTong7 · 2026-10-13
- 13B model trained only on pre-1931 text shows personas emerge from training data era — sethlazar · 2026-10-13