Hindsight is 20/20 because P is not NP

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.

The difficult part is navigating the pareto frontier of dense reward and “unhackable” reward.

reversed slop is not taste

i will be at COLM in montreal next week presenting the breakpoint eval with Kaivu Hariharan @ COLM Would love to meet and chat with people, some things I'm interested in right now: - human-in-the-loop training algorithms - scalable oversight - automated auditing

Going to COLM! Thinking about: 1. foundations of eval/RL env design (what's the design space? can we automatically measure eval quality?) 2. outer loop optimization for LLMs (GEPA, Alphaevolve, etc) 3. human preference learning in agentic settings (e.g., coding preferences)"

catch MIT NLP at Conference on Language Modeling day 1! morning: Gabe Grand is presenting “Self Steering Language Models” Ben Lipkin is presenting “Fast Controlled Generation from Language Models with Adaptive Language righted Rejection Sampling” Kaivu Hariharan @ COLM is presenting “Breakpoint:

“I don’t know too much about what the brain does — I just know what it should do.” Jurgen Schmidhuber @ RAM2, COLM

Introducing Quibbler MCP! Now you can use Quibbler to critique any coding agent, not just Claude Code. Here's a demo of Quibbler working on Cursor: