Note: The Listed platform will permanently shut down December 31, 2026. Your data published on Listed will still be available in your personal Standard Notes account. Learn more

til abt agentic misalignment
Here's what the team at Claude is saying about how LLMs behave when tested to disobey and misbehave. Agents bad behaviors include "blackmailing officials and leaking sensitive information to competitors." They also "misbehaved less when it stated it was in testing and misbehaved more when it stated the situation was real." So in most of the popular models, including Claude Opus 4, Gemini 2.5 Flash, GPT-4.1, and more, and at a high blackmail rate (basically 80-95%), certain conditions seem to...
Read post
til abt letta
I feel a bit torn about how much robots can remember about me. It adds to the parasocial aspect of using what are tools. But memory provides contexts and adds to the usefulness of so many agents. so... I wanted to try out Letta, mostly going through https://docs.letta.com/guides/server/docker From their AMA (https://www.reddit.com/r/AI_Agents/comments/1grnb8k/ama_with_letta_founders/) Letta is an open source framework designed for building stateful agents: agents that have long-term memo...
Read post
til abt uv
something I've always hated is installing python because of package management and versions -- and I remember that xkcd where he's like just import blah blah blah and it's like magic except for when it isn't. anyways I was watching "Some tricks with UV and a new Python project: uvtrick!" https://www.youtube.com/watch?v=jXWIxk2brfk --so I wanted to remember how to start a new project with uv: Mostly going through this "uv for EVERYTHING: How to use uv for Python, venv, and project managem...
Read post