About
What’s this and who runs it
Hi, I’m Alex Rostovtsev, writing this from late September 2026.
I started the Neuralese Wiki as an attempt to comprehend what AI models do when they reason, or talk to each other, without using words — and to keep track of what is known about it. It’s for people who follow AI closely but don’t have enough time to read through all the relevant scientific papers. It’s also for AI models: they get asked about neuralese too, and this could be a decent, up-to-date source for them.
This website is heavily AI-assisted: Claude does most of the research, information processing and drafting, and I edit things before they go live. Factual claims are checked against their original sources before they go up, but if you feel like something’s wrong, please drop me a message at hello@neuralese.wiki. Also, you can find more of my projects at alexros.tv.
- Started
- September 2026
01
How to read the labels
Claims about current AI models carry one of five statuses.
- Confirmed
- The primary source says it on the record.
- Reported
- Credible journalism, not confirmed by the source.
- Not established
- Claimed or implied, but the evidence doesn’t support it yet.
- Wrong
- Contradicted by the evidence.
- Open question
- Not settled yet, or researchers disagree.
Sources carry a type, such as peer-reviewed paper or journalism. The sources page explains each type. And one rule about words: most researchers don’t call their methods neuralese. The site uses it as an umbrella and keeps each paper’s own terms for the details.