Self-taught 18-year-old publishes audited study on moral/emotional tone steering in LLMs
perseverance_remains · reddit · 2026-10-10
An 18-year-old self-taught ML researcher published an audited study (paper + code, moralengine) on moral/emotional tone steering in LLMs. Key findings: a proof that probe-derived steering keys are mathematically confined to their own row space (∇f(x) ∈ rowspace(W₁)); reading works, with grouped MLPs scoring 0.78–0.98 on template-grouped CV; but steering verdicts show Care/Harm is null while Arousal is a candidate (1 of 2 seeds pass). The author seeks expert feedback on methodology, the rowspace proof, and next experiments, and is open to collaboration.
More from Research
- EMNLP paper: diffusion LMs internally encode latent denoising timestep representations — gsarti_ · 2026-10-23
- KU Leuven and VIB open two Group Leader professorships in AI for Biology, start 2027 — jpillowtime · 2026-10-23
- Subspace Tuning curbs LLM catastrophic forgetting by freezing principal singular directions — scottniekum · 2026-10-23
- Perceptic launches PercepticOS, an AI workspace for 'scientific superintelligence' — nathanbenaich · 2026-10-23
- EPFL's Bosselut to talk long context as test-time learning at ELLIS seminar — andre_t_martins · 2026-10-23
- DRATS: Adaptive Task Sampling Boosts Worst-Task Performance in Multi-Task RL — Cohere · 2026-10-23