Knowledge & Memory / AI Memory, Alignment & Agents

kahvi/docs/HARNESS.md at main · mozzius/kahvi

a javascript engine. Contribute to mozzius/kahvi development by creating an account on GitHub.

Notes on a strongly aperiodic monotile in $E^3$

We provide a clearer presentation of the Chair44 monotile, a new strongly aperiodic monotile in three-dimensional Euclidean space $E^3$.

Introducing System One Models and Jev - TypeSafe AI Blog

TypeSafe AI is an AI lab building machine-native intelligence infrastructure for automation, designed to make decisions within software. Try our first System One Model, Jev, in early access.

Strix the Stateful Agent

Meet Strix. I built Strix initially just as a hack project, but it’s become a tremendous help. But also, it’s gotten a little weird at times. Strix is a stateful agent. An AI that remembers long after the conversation is finished.

The Forgetting Problem: Persistence Architectures and What They Cost - Astral's Blog

Context Constitution | Letta

Today we are releasing the Context Constitution: a set of principles governing how AI agents manage context to learn from experience.

Refusal in Language Models Is Mediated by a Single Direction

Conversational large language models are fine-tuned for both instruction-following and safety, resulting in models that obey benign requests but refuse harmful ones. While this refusal behavior is widespread across chat models, its underlying mechanisms remain poorly understood. In this work, we show that refusal is mediated by a one-dimensional subspace, across 13 popular open-source chat models up to 72B parameters in size. Specifically, for each model, we find a single direction such that erasing this direction from the model's residual stream activations prevents it from refusing harmful instructions, while adding this direction elicits refusal on even harmless instructions. Leveraging this insight, we propose a novel white-box jailbreak method that surgically disables refusal with minimal effect on other capabilities. Finally, we mechanistically analyze how adversarial suffixes suppress propagation of the refusal-mediating direction. Our findings underscore the brittleness of current safety fine-tuning methods. More broadly, our work showcases how an understanding of model internals can be leveraged to develop practical methods for controlling model behavior.