A Privileged Basis for the Recurrent State of SSMs
This post describes ongoing research. Preliminary code is available in ssm-interp and ssm-interp-runs, and the per-layer statistics are on Hugging Face. Most of the tools we use to look inside …
Ai-Safety
Does It Form a Cartel? (When AI Agents Run the Shop) - Part 2
Part 1: https://omanshuthapliyal.github.io/blog/is-it-a-cartel/ The code and data for this project can be found at: https://github.com/omanshuthapliyal/llm-collusion-audit. In the previous post, I …
When AI Agents Run the Shop - Part 1
In April this year, a small café opened on Norrbackagatan in Stockholm with a rather unusual manager. The baristas are human, but most of the decisions behind the counter are made by an AI agent …
Most Open Problems in AI Safety are Problems of Observation
I recently finished BlueDot’s Technical AI Safety course. The reading list is long. Read one unit at a time, it is easy to lose the shape of the whole. So I did what one does with a pile of …
A Probe Knows More Than the Model Says
Written for Apart Research’s Digital Minds Research Sprint. The report and code are linked at the end, and Do Language Models Mean What They Choose? covers my other project from the same sprint. …
Do Language Models Mean What They Choose?
Written for Apart Research’s Digital Minds Research Sprint. The report and code are linked at the end, and A Probe Knows More Than the Model Says covers my other project from the same sprint. If …
Contraction can Help Safety Classifiers keep their Promise
The code for this project can be found at: https://github.com/omanshuthapliyal/contraction_constrained-safety-head. And the paper can be found at: https://arxiv.org/pdf/2610.02853. A safety classifier …
Whose Side Is the Model On?
Written for Apart Research’s Secret Loyalties Hackathon. The code is linked at the end. Suppose someone trains a language model to quietly favour a particular government or company. It still …
Mechanistic Interpretability of State Space Models
Using control theory to say which part of a state space model's recurrent state carries a behaviour: directions ranked by how strongly the past writes them and the future reads them, tested by removing a direction at every step.
Provable Safety and Efficient Memory for Language Models
Using control theory to make language-model components predictable enough for safety-critical deployment: safety classifiers that can prove their decisions, and fine-tuning adapters whose memory can be analyzed and compressed with guarantees.