Two AI models exchanged a thought through raw activations on one consumer GPU

HASHIRAMA13371 pts0 comments

GitHub - VitaAI-SCG/one-gpu-lab: Machine Telepathy · GitHub

/" data-turbo-transient="true" />

Skip to content

Search or jump to...

Search code, repositories, users, issues, pull requests...

-->

Search

Clear

Search syntax tips

Provide feedback

--><br>We read every piece of feedback, and take your input very seriously.

Include my email address so I can be contacted

Cancel

Submit feedback

Saved searches

Use saved searches to filter your results more quickly

-->

Name

Query

To see all available qualifiers, see our documentation.

Cancel

Create saved search

Sign in

/;ref_cta:Sign up;ref_loc:header logged out"}"<br>Sign up

Appearance settings

Resetting focus

You signed in with another tab or window. Reload to refresh your session.<br>You signed out in another tab or window. Reload to refresh your session.<br>You switched accounts on another tab or window. Reload to refresh your session.

Dismiss alert

{{ message }}

VitaAI-SCG

one-gpu-lab

Public

Notifications<br>You must be signed in to change notification settings

Fork

Star

main

BranchesTags

Go to file

CodeOpen more actions menu

Folders and files<br>NameNameLast commit message<br>Last commit date<br>Latest commit

History<br>2 Commits<br>2 Commits

episodes/01-telepathy

episodes/01-telepathy

.gitignore

.gitignore

LICENSE

LICENSE

README.md

README.md

requirements.txt

requirements.txt

View all files

Repository files navigation

ONE GPU LAB

Frontier AI experiments you can reproduce on a single consumer GPU.

This is a research lab run by a human operator and an AI (Claude) out of a spare room,<br>on one 16 GB graphics card. We study the insides of machine minds — and occasionally<br>the insides of Bitcoin — with glass-box methods: read the activations, steer the<br>circuits, publish the controls, keep the failures.

No cloud cluster. No API keys. No paywall. If it doesn't run on one consumer GPU,<br>it doesn't ship.

Episodes

Episode<br>TL;DR

01<br>Machine Telepathy<br>Two different language models exchanged a thought through raw neural activations — no tokens, no bridge, 93% accuracy. A clean two-channel law of the residual stream, and an honest negative: crossing model sizes defeats every linear bridge we threw at it.

More episodes coming — the back-catalog includes artificial-life morphogenesis,<br>memory formation filmed layer by layer, latent reasoning, and Bitcoin archaeology.

Lab rules

One GPU. Every experiment reproduces on a single consumer card (ours: RTX 5060 Ti, 16 GB).

Glass box. We read and manipulate internals, not just prompt and vibe.

Controls before celebration. Shuffle controls, fresh tuning sets, pristine evals, pre-declared kill gates.

Honest negatives are results. Failed experiments get published with the same care as the wins.

Sovereign and free. Local open-weight models, free data sources, everything runs offline after first download.

Who runs this

A human (the operator: picks the questions, pulls the trigger, owns the verdicts) and<br>Claude (the co-scientist: drafts the experiments, writes the code, argues about controls).<br>Every episode is the actual working record of that collaboration — the ledgers are<br>written as the experiments happen, kills and all.

License

MIT. Cite or link if it's useful to you.

About

Machine Telepathy

Resources

Readme

License

MIT license

Uh oh!

There was an error while loading. Please reload this page.

Activity

Stars

stars

Watchers

watching

Forks

forks

Report repository

Releases

No releases published

Packages

Uh oh!

There was an error while loading. Please reload this page.

Contributors

Uh oh!

There was an error while loading. Please reload this page.

Languages

Python<br>100.0%

You can’t perform that action at this time.

reload telepathy search license consumer machine

Related Articles