blank

Split conformal prediction has become a popular frequentist tool for putting uncertainty on a model’s predictions.

AGI is everywhere in the news right now, although in my view we have not achieved it yet.

Forecasting is usually treated as prediction with the target moved further out, which quietly assumes the future looks like the past.

Each time a frontier model ships, one camp gets more confident AGI is close and the other gets more confident it isn’t, both watching the exact same release.

Picture two engineers shipping the identical feature.

Peter Thiel’s claim about distribution is usually read as a sales lesson.

“Minimum viable product” is one of the most misread terms in product development.

A production incident gets fixed under pressure late on a Friday, and by Monday the engineer who saved the day is the center of attention.

Everything we can imagine is assembled out of things we have already perceived.

Today’s AI has no continuing self.

Research and engineering are equally demanding disciplines when done right.

Introduction Bayesian inference provides one of the most principled approaches to learning from data.

Uncertainty estimation is often framed as the key to building reliable machine learning systems.