Ensuring AI alignment research keeps pace.
As soft takeoff picks up, Alignment Hive empowers the third-party alignment research community to keep up through curated tools, training, and aggregated knowledge and data: the benefits of scale that frontier AI labs have.
a step function change in my research productivity. good to get 1) a productivity reflection and perspective with an expert like Yoav who's also familiar with eng research workflow 2) up-to-date on the latest agentic features
Fully using all the features while trying to hail Mary two first-author papers into COLM, and I'm just wondering how I ever let myself go without it. Alignment Hive is so good, thank you so much, Yoav
Thanks for the Claude Code setup boost. It's so much fun talking and sharing workflows, especially when the advice comes from someone who has put in enough hours and not just hyping things up like influencers.
I very strongly recommend attending a session with Yoav. My workflow has leveled up massively since our session -- I just wish I had signed up sooner!
Yoav was uniquely helpful at MATS. He helped me make real technical headway in just a couple meetings, and ultimately led me to reconsider the mechanics underlying my core project.
Talks, workshops, & consulting
Talks, workshops, and consulting for AI safety fellowships and research orgs: where the state of the art of reliable agent delegation lies, tools and techniques that help you get there, and personalized hands-on advice.
Email meTools
Open source and shaped by concrete bottlenecks in real research work. Here are some of the most-used tools in the collection:
Hive
Alignment Hive entry point and memory for Claude Code: instead of relying on model-written memories, Claude searches your past sessions directly. Also home to the community's opt-in knowledge and data aggregation.
Codex for Claude Code
Codex reviews and task delegation without leaving Claude Code. Claude understands fuzzy constraints better; GPT is more careful and catches mistakes; together they're more than the sum of their parts.
Remote Kernels
Unblock Claude to run experiments for you autonomously by letting it operate cloud compute efficiently and reliably.
Used by dozens of researchers across MATS, Far Labs, and other AI safety organizations.
About
Alignment Hive is built by Yoav Tzfati: MATS alum (scalable oversight with NYU's Alignment Research Group, then the Security Level 5 task force). Previously built the infrastructure for the Arbor trading bootcamp and helped teach it, and taught Claude Code at Code Bloom just four months after its public release.