Hyacehila's Blog

Starting from the view that loss is only a surrogate, this post revisits test-time compute, capability frontiers, policy entropy collapse, and Meta’s Three-Gate theory.

Based on Unveiling the Basin-Like Loss Landscape in Large Language Models, this post explains basin-like loss landscapes and implications for fine-tuning and alignment.

A summary of Jack Rae’s Compression for AGI: foundation-model training as lossless compression of useful information, and why lower loss can imply stronger generalization.

Is the Voynich manuscript random scribbling or a lost language? What would alien signals look like? Both questions point to the same statistical law.

A repost of Jeff Leek’s classic guide to preparing and organizing data before working with a statistician, covering raw data, tidy data, and code books.

An introduction to MCP’s Host, Client, and Server layers, stdio and Streamable HTTP transports, and the workflow for building and debugging MCP servers and clients.

A discussion of why statistics drifted away from applied settings while pursuing theoretical completeness, and how data science and machine learning reconnect with learning from data.

A review of 60 years of data science, from John Tukey’s prediction to David Donoho’s broader view, discussing how statistics, computer science, and machine learning converged.

R-squared is not a simple model error function. Its definition hides a comparison against a constant model, so it never directly measures fit to the real world.

A repost of Jonas Kristoffer Lindelov’s classic article showing the unified linear-model logic behind t-tests, ANOVA, chi-square tests, and other common methods.