Introducing ProvableWorldModel: the first provable JEPA world model. A real pretrained world model runs, and anyone can verify, on a laptop CPU, or on a mobile phone, that the committed model ran exactly as claimed, in milliseconds. The whole industry spent this week asking one question about Anthropic's Fable 5: which model actually answered me? Nobody could prove it. Not during the silent downgrades, not after the apology. The fix is a notification you still have to trust. I've been building the alternative. The intuition is simple. World models are small. Small enough to run in exact integer arithmetic, where every operation is reproducible bit for bit, attention included. The prover commits to its full execution trace before any challenge is derived. The verifier fingerprints the heavy matmuls with 1979-era randomized algebra (Freivalds) and replays everything else exactly. Change one number anywhere and the proof dies. Receipts: → real pretrained JEPA predictor (LeWorldModel), 10 MiB int8, real expert episode → inference 49 ms, verification 29 ms, CPU only, no floats, no GPU → soundness error ≈ 2^-44 → forge a single matmul output and the verifier throws FreivaldsCheckFailed Why it matters: LLM agents need ever-thicker invisible harnesses to behave, and harnesses can't be audited from the outside. World models invert that. Small, structured, exact: verifiable by construction. Intelligence you can check, not intelligence you have to trust. Don't trust, verify.

Using Fable and watching your usage bar evaporate

The top two social apps on @ethereum are on Starknet

we received a lot of love for this launch and we wanna give back to you, yes you. i'm giving away 200 of these very limited tees/ caps to people who RT the main post (comment proof here) taking a snapshot in 24 hours.

Privacy matters. This may make tracking certain metrics more challenging in the future, but we welcome that challenge and will never stand in the way of progress. Progress is happening on @Starknet

Introducing: Proof. Starknet’s selective incubator and accelerator for high-potential, investor-grade teams. Applications for Cohort 01 are now open.

1/ JUST IN: practical privacy at real scale is now live. 1-click privacy, low-cost privacy, private DeFi… For all assets, now live on Starknet 🧵 strk20.starknet.io/

This is the playbook. It works because life is long and memories are short

1/ introducing zylith a call auction dark pool that enables private order flow while reducing timing and access-pattern leakage private orders enter fixed auction epochs, are matched against aggregated demand and supply, and clear at a single uniform price

a complete masks off moment for ai safety. anthropic wants to be the only lab. it hates its users and humanity. it won't allow random users to do defensive security work or understand diseases. it will silently poison your prompt if you try to develop ml code. the only feeling they know is contempt.

Privacy finally found the right frequency. [STRK20] is live on Starknet. 🧵

The @Zcash bug leads to Ironwood, formal verification of the Orchard payment circuit. I love and support it. I also want to stress that @Starknet has been doing that -- formal verification -- for over 5 years. It's important to support great projects like Zcash for reacting quickly, as in this case. It’s also important to recognize projects that are future-proof and ahead of the curve, like Starknet. (Details in the article below) I’ve said often that Starknet is that system which already has the stuff other chains claim they soon will. It's been true for many things. We've been at the vanguard of many unpopular choices that now everyone recognizes are important, including post-quantum secure ZK-STARKs as the best scaling and privacy solution, lean zkVMs (Cairo is best), Validium data availability, and formal verification. Formal verification means that you use automated tools like the Lean system to mathematically prove that your code is safe. It's very hard to capture each and every aspect of what it means for code to be safe, but led by our CTO @LiorGoldberg2 (co-creator of Cairo zkVM and language), we've been at it for more than 5 years. The very first paper on formal verification of claims related to ZK (Professor Jeremy Avigad, Yoav Seginer, and others) showed that the set of polynomial constraints defining the Cairo VM is correct. If you've been following the news lately, the bug that was recently discovered by AI and now fixed for Zcash (another project I co-founded, and which I'm very proud of and support) had to do with a missing constraint. The decision to verify their code to rule out other such bugs is the right step. Running this kind of verification for the Cairo core VM provided us, the StarkWare team, with mathematical certainty that there is no missing constraint in our Cairo VM. So, in addition to proving the core VM of Cairo, we have also proved the S-two STARK system, part of the compiler, and many of the functions in the standard Cairo library. Is the job done? No. There are a lot of other things that can be proven formally . But we are committed to continuing to formally prove all the core properties that have to do with the soundness and safety of our systems. It's great to see other projects embracing formal verification as an important tool. I'm proud that StarkWare led the path in this aspect too. That’s what Future Proof blockchain means to us. Want to learn more? Here's a detailed account of our Lean proving efforts over the past few years

We made a @UnitreeRobotics Go2 leaves receipts for its motion decisions Before each Move, WorldForge checks whether the command should still be allowed A marked suitcase enters the path. The move is vetoed. The robot stops Path clears. The Move loop resumes Every tick emits a DecisionTrace and SHA We used direct Unitree WebRTC. Next step is wiring the same loop through the @dimensionalos Go2 adapter (Im sure it will be much easier and the dog will actually do interesting moves) github.com/AbdelStark/worldf…