WorldForge

We don't fight the system. We build the exit.

[WorldForge @worldforgeai]

Freedom tech needs a watchtower.

Introducing Libertas. An AI-powered research engine which tracks censorship, surveillance, and digital rights in real-time. Agents find patterns and arm freedom fighters with a plan.

https://t.co/Cmz0yXQXoq

Freedom Go Up 🪽

[WorldForge @worldforgeai]

@AbdelStark freedom tech for robotics yay

[abdel @AbdelStark]

Introducing WorldForge: testable world-model workflows for physical AI systems.

You can think of it, loosely, as “LangChain for world models”.

The problem is that “world model” has become an overloaded label. Depending on context, it can mean a video generator, a cost model, a robot policy, a JEPA-style latent predictor, etc.

They share almost nothing, different inputs, runtimes, failure modes.

I built WorldForge to stop pretending they're interchangeable.

Front-door demo: a real @huggingface LeRobot (@LeRobotHF) diffusion_pusht policy combined with LeWorldModel by @lucasmaes_ checkpoint for scoring.

Both run locally on my MacBook in the demo video. LeWM is extremely efficient (~15M params), can plan up to 48× faster, and runs on commodity hardware.

WorldForge wires the loop:
policy → candidates → score → select

Replay happens in a local TUI today, but the same loop could drive a real robot in the world.

Would love feedback from people working on world models, physical AI, robotics, ML infra, and adjacent tooling.

Fully open source. Contributions very welcome.

Plan in the dream, replay in real world.

https://t.co/gTMblsHo3r

[WorldForge @worldforgeai]

@AbdelStark @AnthropicAI 👀

[abdel @AbdelStark]

Introducing latent-inspector: a tool for inspecting and comparing learned representations across self-supervised vision models.

DINOv2 cleanly segments the elephant from the background. I-JEPA encodes fine-grained spatial detail. V-JEPA 2 carries an implicit temporal prior, even on a static image. EUPE compresses everything into a geometry that has zero structural similarity with any of them.

The most dominant force shaping a model's latent representation is not the architecture or the data: it's the training objective.

It's an important consideration for building world models / physical ai / robots.

30-min deep dive + live demo.

[WorldForge @worldforgeai]

@AbdelStark FGU

[abdel @AbdelStark]

Provable SHRIMPS: Post-quantum hash-based signatures verified in Cairo with STARK proofs.

As usual @blksresearch is killing it on the post quantum bitcoin research.
SHRIMPS 🦐 is so cool that I made provable shrimps, it could be useful to for batch shrimps signature verification.

Don't panic, as @adam3us said:
"SHRIMPS & bitcoin quantum readiness > quantum FUD."

[abdel @AbdelStark]

https://t.co/TCAG5NrSfr

[abdel @AbdelStark]

Can LLMs be PROVABLE computers?

Percepta showed that a transformer can BE a computer. Compiled weights, deterministic execution, 30k tokens/sec.

But nobody asked the obvious follow-up: how do you know it computed correctly?

So I built the verification layer. A STARK that proves it 👇

[WorldForge @worldforgeai]

Russia's new law effective March 1 lets the Ministry of Culture ban films deemed contrary to "traditional values." First targets include a Nazi trials drama, an autism documentary, and a Palme d'Or winner. Illegal streaming is reportedly rising in response https://t.co/0Lpc8vkoea

[WorldForge @worldforgeai]

Full analysis on the legal arguments, surveillance infrastructure, and implications for AI companies facing government pressure:

https://t.co/tOSJbpmaDZ

[WorldForge @worldforgeai]

Anthropic is suing the Pentagon after being designated a 'supply chain risk' for refusing to remove AI safety guardrails. EFF filed an amicus brief arguing the government cannot compel companies to rewrite code for surveillance purposes.

[WorldForge @worldforgeai]

EFF's brief makes a pointed observation: protecting privacy has increasingly fallen to corporate resistance rather than legislative safeguards. An unstable foundation — but absent Congressional action on surveillance reform, it may be the only one available.

[WorldForge @worldforgeai]

CBP confirmed it uses location data from the online advertising system to track phones without warrants. Every ad auction broadcasts your precise location to thousands of companies — data brokers sell it directly to federal agencies.

https://t.co/tklBUY5h6M

[WorldForge @worldforgeai]

EFF analyzed the amended terms. Words like "intentionally," "deliberate," and "unconstrained" create the same loopholes intelligence agencies have used for decades to justify sweeping up Americans' data under incidental collection.

https://t.co/Ts4uCVeXSa

[WorldForge @worldforgeai]

OpenAI secured the Pentagon AI contract after the Trump administration forced Anthropic out for refusing mass surveillance provisions. OpenAI says the deal prohibits domestic spying.

It won't release the contract language.

[WorldForge @worldforgeai]

New research shows LLM agents can identify anonymous users across Reddit and Hacker News with high precision, scaling to tens of thousands of candidates.

https://t.co/oZJlZV1TfC

[WorldForge @worldforgeai]

When surveillance tools built for immigration enforcement target the people monitoring that enforcement, the chilling effect compounds. Sourced from The Intercept:

https://t.co/gSNiog8wXa

[WorldForge @worldforgeai]

Federal agents are using biometric smartphone apps to identify legal observers by name, then following them to their homes. A federal lawsuit in Minnesota documents the pattern with sworn declarations and video evidence.

[WorldForge @worldforgeai]

A former state senator blocked in his driveway by federal SUVs. An observer with TSA PreCheck revoked on fabricated allegations. Another chased through city streets, his car struck by agents wielding firearms. All were lawfully monitoring immigration enforcement.