New Paper TrustMI Finds a Linear "Trust Direction" in 6 LLMs
ThomasScialom · x · 2026-10-14
Researchers released a new paper, TrustMI, asking a simple question: does your AI model trust you?
- They identify a linear "trust direction" in six large language models;
- Both the paper and a Hugging Face page are public;
- Advisor zehavoc shared the work proudly, noting a demo video is included.
It sits in the LLM interpretability and internal-representation line of research.
Related event: TrustMI paper causally steers how much LLMs trust users(3 posts)→
More from Research
- Open-Science 1.0 launches as open-source AI workbench bundling research workflows into reproducible .science packages — alifcoder · 2026-10-15
- Decoding NCM's latents reveals coarse-to-fine text generation trajectory — Azaliamirh · 2026-10-15
- Frontier AI models now win 90%+ of Slay the Spire runs, up from barely winning six months ago — NathanpmYoung · 2026-10-15
- Nvidia and ChipAgents launch Build-A-Chip-Bench; top AI agent scores just 16.6% — teortaxesTex · 2026-10-15
- 108 OLMo-style LMs (4M–75M) with ~3.9K checkpoints released from ICML paper — BrihiJ · 2026-10-15
- Google's DuoDiscover tops NVIDIA SOL-ExecBench, 2.68x faster than production TPU kernels — lateinteraction · 2026-10-15