Blog

Practical writing on AI engineering — from research to production.

One practical tip on how to talk to your AI, with receipts

I gave six reasoning models the same questions with explicit names and with pronouns, then compared their reported thinking tokens.

machine learningbenchmarksreasoninglanguage models

Another easy way to break LLMs I found

Check if you can beat LLMs at finding a hidden curve from the empty space in a point cloud.

machine learningbenchmarksvisual reasoningmathematics

Simple Stats intuition AI fails on

Pick which of those two was generated completely randomly

Statistics Intuitionsstatisticsprobabilitymachine learning

Do you know when you know: Testing LLMs for their metacognitive skills

Check if you can beat LLMs at these mathematical visual reasoning tasks.

machine learningbenchmarksmetacognitionvisual reasoning

Retargeting Controller

Control a simulated robot arm with your hand through a mapping you can rewire.

side-projectroboticssimulation

Check your intuition

Which of those two patterns was generated completely randomly?

Statistics Intuitionsstatisticsprobability