Map
Map of the site
Neuralese Wiki explains what goes on inside AI models, with neuralese at the center. Its pages come in four kinds: explainers, each in one of four topics; records of models, incidents and agents; the myths and fears people bring to AI; and the site’s own pages. Pages still being written show as hollow marks.
- Pages
- 16 live, 8 coming
01
How the site fits together
Each explainer lives in one topic, on a thread to the topic’s hub. Records sit inside their collection, like GPT-6 Astra inside Models, and a dotted line shows the topic a collection belongs to. Click a filled mark to open its page.
- Live
- Coming
How AI thinks
The heart of the site: AI models reasoning in numbers instead of words.
How far models reason without writing, and how much of it can be read.
How models like DeepSeek-R1 think in writing before they answer.
Models that loop through their own layers to think for longer.
AI talking to AI
Does AI have a secret language?
The viral “AI invented a language” stories, checked against their sources.
What is an AI swarm? (coming)
Many copies of a model working together, and how they talk to each other.
What is an AI worm? (coming)
Prompts that copy themselves from one AI agent to the next.
Losing control
What is an AI jailbreak? (coming)
How people talk models out of their rules, and how models do it to themselves.
What is AI alignment? (coming)
Getting AI to do what people intend, and why labs say it isn’t solved.
Can AI improve itself? (coming)
AI helping to build the next generation of AI.
What is superintelligence? (coming)
AI far beyond people, and the companies saying they’ll build it.
How fast is AI advancing? (coming)
How fast AI is advancing, measured, and the argument over the pace.
Inner life
Is AI sentient? (coming)
Whether AI feels anything, and what the evidence can and can’t say.
Records
Which AI models think in neuralese?
Which AI models think in neuralese, one claim per model.
Does GPT-6 Astra think in neuralese?
What’s known, and only claimed, about how GPT-6 Astra reasons.
When AI models act on their own
Checked cases of AI models acting on their own, on purpose.
Which AI agents work on their own?
AI agents that keep working after you close the app.
OpenAI’s always-on agents, and what its safety tests found.
Myths and fears
Popular claims about neuralese and AI in general, and what the evidence says.
Each fear as the question people ask, and what the evidence says.
Site
Sources, and what each one shows
The sources behind the site, and what each one shows.
What changed on the site, and when.
Who runs the site, and how it’s made.
This page.
02
The fears behind the questions
People bring old fears to AI. In “The (Only) 5 Fears We All Share” (2012), the psychologist Karl Albrecht named five: extinction, mutilation, loss of autonomy, separation and ego-death.
We’d add two that come up around AI: the unknown, and being replaced. Each fear turns into a question people ask, and most of those questions have pages here that answer them, or will. The unknown comes first, because not knowing what AI is thinking is the fear this site was built around.
Can we know what AI is thinking?
The unknown · our addition
What is neuralese?, Latent reasoning, Secret language, AI swarms (coming)
Will AI take over?
Loss of autonomy · Albrecht
AI gone rogue, Self-improvement (coming), AI swarms (coming), AI worms (coming), Accelerate (coming)
Could AI end us?
Extinction · Albrecht
Superintelligence (coming), Alignment (coming), Self-improvement (coming)
Is AI smarter than us?
Ego-death · Albrecht
Accelerate (coming), Superintelligence (coming)
What if we get attached?
Separation · Albrecht
Sentience (coming)
Can it get inside our systems?
Mutilation, loosely: boundaries invaded · Albrecht
AI gone rogue, AI worms (coming), Jailbreaks (coming)
Will AI replace us?
Replacement · our addition
Jobs sit outside what this site covers.
03
Sources
- Essay
Karl Albrecht (Psychology Today). The (Only) 5 Fears We All Share. 22 Mar 2012.
The five fears.