nader dabit

the cerebras post is really good, i have been feeling the urge to go build an agent i was gonna call “librarian” to just index and dedupe slack threads i find myself often redirecting people to where conversations are already happening on x topic since i read all threads anyway

Cognition makes over $492 million in ARR. Yet, I don't know a single person who use Devin. What's going on here?

1,000 tok/s vs 85 tok/s visualized

Introducing SWE-1.7, the most capable model we’ve trained yet. It scores within a few points of the strongest frontier models at a fraction of the cost, and is now available at 1000 tok/s. RL is not hitting its limit: after refining our recipe, we keep seeing gains as we scale

Introducing SWE-1.7, the most capable model we’ve trained yet. It scores within a few points of the strongest frontier models at a fraction of the cost, and is now available at 1000 tok/s. RL is not hitting its limit: after refining our recipe, we keep seeing gains as we scale

Introducing SWE-1.7, the most capable model we’ve trained yet. It scores within a few points of the strongest frontier models at a fraction of the cost, and is now available at 1000 tok/s. RL is not hitting its limit: after refining our recipe, we keep seeing gains as we scale

1,000 tok/s vs 85 tok/s visualized

1,000 tok/s vs 85 tok/s visualized

Introducing SWE-1.7, the most capable model we’ve trained yet. It scores within a few points of the strongest frontier models at a fraction of the cost, and is now available at 1000 tok/s. RL is not hitting its limit: after refining our recipe, we keep seeing gains as we scale