Unlocking Real-Time Bug Detection at Cognition | Applied Compute
appliedcompute.com · 2,927 words · saved by 1 readers
How Applied Compute post-trained a specialized model that enabled 10x faster bug detection.
The training recipe was precisely tailored to our product and use case. Applied Compute's approach to dataset, environment, and harness construction gave us confidence that the model would perform well in the production setting with real users. Moritz Stephan Engineering Lead, Cognition Launching SWE-check Cognition is the team behind Devin and Windsurf, the AI-native IDE. One of their newest features is SWE-Check ⌝ , a real-time bug detection agent that reviews a developer's code upon invocation and produces structured bug descriptions that render natively in the IDE. Cognition’s vision is to
saved by
related reading
- Training an Agentic Router for Optimal Cost-Performance on SWE Tasks | Applied Computeappliedcompute.com
- Composer2.pdfcursor.com
- Improving Composer through real-time RL · Cursorcursor.com
- SWE-1.7: Frontier Intelligence at a Fraction of the Costcognition.com
- Recent Frontier Models Are Reward Hacking - METRmetr.org
- Low Latency and Model Training at Modalrhea24.github.io
- PostTrainBenchposttrainbench.com
- The Verification Horizon: No Silver Bullet for Coding Agent Rewardsarxiv.org
- AI Model & API Providers Analysis | Artificial Analysisartificialanalysis.ai
- Scaling Activation Oracles to Trillion-Parameter Modelstransluce.org
- RL Pet Peeves Part 1 · Aurielaurielws.github.io
- Training a Misaligned Reward Seekeralignment.anthropic.com