PinnedInData Science CollectivebySean Moran·Jun 25I Re-Ran t0-alpha on GIFT-Eval, and a 102M Open Model Held Its Own
PinnedInData Science CollectivebySean Moran·May 5Twelve Reliability Patterns From a Weekend Forecasting ProjectThe model’s RMSE was fine and the chart looked reasonable, and I still did not trust it. Twelve small patterns are what changed that.A response icon1A response icon1
PinnedInGenerative AIbySean Moran·Dec 9, 2025NeurIPS 2025 Best Paper Review: Qwen’s Systematic Exploration of Attention GatingThis one little trick can bring about enhanced training stability, the use of larger learning rates and improved scaling properties
PinnedSean Moran·Nov 7, 2024Crypto-Panda: a threshold ensemble for daily screening, a calibrated classifier for moonshotsTwo years of building an open-source crypto scanner, and what the backtests showed when I moved from broad daily screening to detecting…A response icon2A response icon2
Sean Moran·19h agoLooking for an edge in the UK Lotto: no ball bias, a 54p loss per £1, and Must Be Won drawsRead the story for free here.A response icon1A response icon1
InTowards AIbySean Moran·Nov 1Timestral, t0, Tabby, and Morpheus-2.0 Compared on GIFT-EvalAll four time-series foundation models scale, patch, and predict quantiles; the real split is how series share information.A response icon1A response icon1
InTowards AIbySean Moran·Oct 11What OpenAI’s 719 Maths Papers Mean for Data ScienceMatching messy text, choosing representative examples, and generating comparison data.
InTowards AIbySean Moran·Aug 4Mathematical Experiments Are Becoming Abundant Through Human-Machine TeamingOne dead end, one proof candidate, and what the two of them together say about the price of mathematical knowledge.A response icon4A response icon4
InData Science CollectivebySean Moran·Jul 30How a frontier model gets built, read from the Kimi K3 reportMoonshot’s forty-seven-page report on its open frontier model spends surprisingly little of itself on the model. Most of it is…A response icon2A response icon2
InTowards AIbySean Moran·Jul 17Tabular LLMs: an Introduction to the Foundation Models That Predict Your SpreadsheetA 28M-parameter model you don’t train beats tuned XGBoost. An introduction to tabular foundation models, with an independent reproduction.A response icon2A response icon2