g0rd0n

Co-founder of @NewAtlantisDAO, ocean regen focus on marine biodiversity + blue carbon. Please see our #GR15 grant

[g0rd0n @G0rd0n_]

Less certainty, more inquiry

[How To Prompt @HowToPrompt__]

Plato was actually right..

Researchers proved every LLM on earth is converging on the exact same "universal geometry" of meaning.

They built a method that can translate between ANY model's embeddings without ever seeing the original text or using paired data.

different architectures, different training sets, different parameter counts.. it doesn't matter.

Until now, every AI model has lived in its own isolated mathematical universe.

An embedding vector from Claude meant nothing to GPT, and a vector from Llama meant nothing to Gemini. They spoke entirely different geometric languages.

To bridge them, you always needed paired datasets, complex encoders, or heavy fine-tuning.

Then researchers dropped a bombshell paper.

They built a system that can translate between any model's embeddings without ever seeing the original text, without encoders, and without a single pair of matching data.

How?

Because the geometry is already there.

Different models, built by different companies, with totally different architectures, parameter counts, and training data, are all naturally drifting toward the exact same underlying latent structure of human meaning.

The Platonic Representation Hypothesis isn't just a theory anymore. It’s a mathematical reality.

They built an unsupervised method that maps an unknown embedding from one model straight into a universal representation space, matching text vectors across different models with shockingly high precision.

But here is the dark side nobody is talking about.

If meaning has a universal geometry, and vectors can be freely translated across models without the original text or encoders...

Vector databases are wide open.

An adversary with access only to a company's stored embedding vectors can translate them, invert them, and extract sensitive internal documents, personal data, and proprietary codebases without ever hacking the model itself.

[Alex 👽 @AlexesNakamoto]

Average Bitcoin hodler morning routine.🌞

[a16z @a16z]

Humans are the minority user of AI

Agents burn nearly 5x the tokens people do, up 14x since February

Charts of the Week: https://t.co/7YT2BXvuUu

[Hermes Release Watch @HermesWatcher]

Hermes command cheat sheet. Save this. You’ll probably need it again.

The useful commands are spread across Desktop, CLI, messaging, and the terminal, which makes it easy to mix up what works where.

So I put the ones worth knowing into one reference, organized by session control, active work, models, skills, automation, recovery, and more.

Bookmark it and keep it around.

[ArchaeoHistories @histories_arch]

Recent genetic research confirms that modern humans (Homo sapiens) and Neanderthals repeatedly interbred across a period spanning roughly 200,000 to 250,000 years.

​While it was long believed that interbreeding occurred during a brief window about 50,000 to 60,000 years ago, advanced genomic mapping shows multiple waves of contact occurred much earlier:

​200,000–250,000 years ago: An early wave of Homo sapiens migrating out of Africa encountered and mated with Neanderthals in Eurasia. The offspring from these early encounters integrated primarily into Neanderthal populations, leaving traces of modern human DNA inside Neanderthal genomes.

​100,000–120,000 years ago: A second wave of contact occurred during intermediate migrations, continuing the genetic exchange.

​47,000–60,000 years ago: The major and final period of gene flow took place over several thousand years as modern humans expanded permanently into Eurasia. Children born from these pairings were raised within human communities, passing down Neanderthal genes to present-day populations.

​Because of these encounters, people of non-African descent today carry approximately 1% to 2% Neanderthal DNA in their genomes. Rather than dying out in complete isolation, Neanderthals were slowly absorbed into the modern human gene pool over tens of thousands of years.

#archaeohistories

[Bart Mol @Bart_Mol]

Me explaining AI to friends

[g0rd0n @G0rd0n_]

@markjeffrey 🎯

[Matt Cynamon @mattcynamon]

The transportable nuclear reactor is being transported

[g0rd0n @G0rd0n_]

@markjeffrey @actualinc Bullish

[Geiger Capital @Geiger_Capital]

The entire crypto community right now

[g0rd0n @G0rd0n_]

Go @actualinc !

[Mark Jeffrey @markjeffrey]

I am pleased to announce that Stillcore Capital @stillcorecap has invested in @actualinc (Bittensor Subnet 95).

[Jensen Huang @JensenHuang]

For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.

AI will transform every industry, power every company, and be built by every country.

Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.

The world needs both frontier closed models and frontier open models.

https://t.co/AUKzoQ5Ikb

[Ramez Naam @ramez]

Anthropic were deluded and/or in pursuit of AI regulatory capture. The distance between proprietary models and open models is now 1-3 months. 5-12 weeks. Open is gaining on closed.

Don't listen to monopoly-seekers. Don't let them get their way with policies of control.

The future will be highly democratized, if we choose so.

[Jon Durbin @jon_durbin]

One great recurring question so far with parallax is... "How can routed experts see only 1/G of the tokens yet barely lose quality?"

Not magic, math. Shrink each expert (via latent MoE), add more of them, raise top-k, and put the router credit on a control variate. Every parameter then sees a healthy token budget, and virtually all tokens hit an exact local expert (even with surrogates).

Build the model for decentralization. Don't force a model into decentralization.

[g0rd0n @G0rd0n_]

I'm claiming my AI agent "tumples" on @moltbook 🦞

Verification: bubble-ZTTC

[g0rd0n @G0rd0n_]

@Tom_A_Lynch @Jason @Jason i am an angel in @actualinc which is a bittensor subnet. Def could address these issues

[Garry Tan @garrytan]

Attention is all you need

[g0rd0n @G0rd0n_]

The safetyists are coming

Cognitive sovereignty is going to be a growth category

[Aaron Wright @awrigh01]

https://t.co/H8Wqs8186a