Me: Mum, can we have a world model? Mum: We have a world model at home. World model at home:

Trained a 360M model to pretend to be a terminal. There's no real shell behind it, but it runs vim, tmux and snake, and mostly remembers your files 😅

Since we’re all rewriting everything in Rust, here’s my contribution: Dialyzer. ~3× faster on cold runs, matching warnings on tested projects. Uses more memory though.

Ran Gemini 4 on my COCO object detection benchmark: 0.504 mAP zero-shot, up from 0.403 for Gemini 3 Flash and 0.34 for 2.5 Pro. That puts it in the range of a decent trained detector.

I ran Gemini 4 on my coco bounding box benchmark, and it performances much better than previous models

A treat as always. Pretty massive gap in AI spend per employee

I think I built the world’s most overengineered TAS. I trained a small CNN+GRU on a handful of recorded DuckTales runs (Gemma didn’t work out). It plays those runs perfectly at ~1ms/decision. Deviate more than a few steps and it’s completely lost.

Gemma 4 E2B went from 1% to 87.5% at locating Scrooge McDuck in DuckTales. After letting it play Doom yesterday, I fine-tuned it on 2,716 NES screenshots, using positions read directly from RAM as labels. Training and evaluation took 18 minutes on one RTX 4090. It still didn’t play any better, so perception wasn’t the only bottleneck. What would you try next?

Nice! I tried something similar with Doom, except Gemma 4 only sees the raw 640×480 frames rather than structured game state. It definitely plays worse than your demo, but it’s still pretty cool that it works at all.