Today, we’re announcing Fulcrum Research, a startup scaling human oversight.
We are building debuggers that tell you why your agents fail, and what your rewards are truly testing for—the first step toward the inference-time infrastructure required to safely deploy agents.
@zacharynado feels like very few people commenting on this actually listened to the full statement. idk anything about the science but it seems plausible that mRNA vaccines are less resistant to mutations? would be curious to see a refutation.
It’s plausible that much more cognitive labor will soon be done by hybrid human-AI systems. How do we effectively oversee models’ interactions with complex systems over time?
I've been exploring some ideas with @KaivuHariharan@atticuswzf on long horizon legible reasoning — getting models to build explicit world representations we can actually debug and steer.
I am going to the bay 02/26-03/06 and want to meet people thinking about these questions.