Subscribe

Andreas Stuhlmueller

Against RL: The Case for System 2 Learning

People are excited about Deepseek R1 (and O1 and O3). It seems like RL might finally be starting to actually work for LLMs, not just the weak approximation that is RLHF. But for building safe superintelligent systems, RL is at best an intermediate step, and at worst an incredibly unsafe

Scaling your impact: A polynomial model of career growth

In the past I've been confused by level frameworks. They often read to me as "incremental steps towards becoming a manager" or "just do bigger projects over time" or "your level is your years of experience," in a way that doesn'

Introducing Notebooks: A New Way to Analyze Papers in Elicit

We're excited to introduce Notebooks, a new interface to Elicit that revolutionizes the way you analyze and work with research papers. Notebooks let you combine papers from multiple searches, chat with them, create summaries, and conduct systematic literature searches, all on one page. Notebooks enable several new workflows

Elicit and AI Safety

The two main impacts of Elicit on AI Safety are improving epistemics and pioneering process supervision.

Elicit's core bets: Be systematic, transparent, and unbounded

Eventually, there will be many AI assistants for chat, question-answering, search, web browsing, writing, and general computer use. They'll generally aim to be trustworthy, easy-to-use tools for important questions and decisions. What makes Elicit different?

Elicit's mission is to scale up good reasoning

Now that Elicit is a standalone company let's review our mission.

Elicit raises $9 million and becomes a public benefit corporation

We're excited to share that Elicit has raised $9 million in seed funding. Elicit was born at Ought, a non-profit research organization, and is now an independent public benefit corporation. Like Ought, our mission is to scale up good reasoning using machine learning, starting with researchers. Elicit is