Abdelkabir Ouadoukou·2d agoI Built a Self-Improving Loop. The Judge Made It Worse.What a pre-registered ablation taught me about guarded AI loops, and why I’m moving to verifiable code tasks next
NYU Center for Data Science·3d agoA New Tool Gives Researchers Evidence to Back Their AI Training ClaimsA researcher can publish an AI model’s code and results and still leave readers unable to check how the model was actually trained. CDS…
Evgeniy V. Dolgov·4d agoPermutation Test for ML: What Null Hypothesis It Tests and What It Does Not ProveDisclosure: This article was written with AI assistance in structuring and editing. The core concepts, arguments, and code are the author’s…A response icon1A response icon1
Kenneth B.C. Wang·Nov 2I Built an Open Lab That Re-Tests Trading-AI Claims, Starting With My Own Paper75 re-runs, a four-task benchmark, a live LLM trading ledger and rl-pricing 1.0 on PyPI, all open source.
Milan Stanisic·Nov 1Published before the numberAn open-source breast ultrasound pipeline. Not a medical device. No accuracy claim in this release
Petar Kabashki·Oct 25Treat Your AI Experiments Like Product ArtifactsIf you can’t inspect an experiment after it runs, its result is weaker than it needs to be.A response icon1A response icon1
Kenneth B.C. Wang·Oct 225 of 7 Decisions Made on a Mac Failed to Replay on LinuxMy decision records needed a narrow floating-point tolerance to cross operating systems, not a weaker definition of a matching decision.A response icon1A response icon1
Kaustubh Rewankar·Oct 16I Trained the Same Model Twice. Why Did I Get Different Results?You trained a model, noted the accuracy, and ran it again. The number changed. Here is why that happens, how big the change can be, and how…
Adithya Kiran·Oct 10I Got 97% Accuracy. But Was My Machine Learning Model Actually Better?When working on a machine learning project, one of the most satisfying moments is seeing the accuracy improve after making changes to the…
InThe AcademicbyDiogo Ribeiro·Oct 10A Seed Is Not Reproducibilityset.seed(42) fixes the starting state of one random number generator. Seven experiments in R on everything it leaves open.
