I write about research that catches my attention: experiments I run, results I check twice, and the occasional opinion I'm willing to defend. Written down while it's still interesting.
Log
2 entries- 0022026-10-15Mammoth with two cavapoos and a Model YThe plan: drive over Tioga, sleep in the car, and hike Duck Pass trail up to Barney Lake with my two dogs.
- 0012026-10-15About a third of the best coding agents' failures are the benchmark's faultI read every one of the 539 failures of the top five systems on SWE-bench Verified. Many of them are failures of the tests, not of the agents.