Building https://xitter.cf/t.co/6qrAMdfhtG @ClawBankHQ Ex @theqacc @0xPolygon @banklessdao
Ohio
Joined February 2010
- Tweets12.4K
- Following8.6K
- Followers15.2K
- Likes40.5K
Pinned Tweet
The old company scaled by hiring people.
The next company scales through compute and refactoring.
Completely different organisms with different economics.
Science has dark matter.
A NeurIPS paper from Max Planck, Mila and others maps coherent ideas no research community can see, not because they're wrong but because nobody holds the right mix of concepts.
Discovery is about to go nonlinear.
arxiv.org/abs/2603.01092
"No text exists to leak."
New hybrid agents reason in latent space, so the thought never becomes a string. Nothing to exfiltrate.
The next era of security isn't encryption. It's minds that think in a medium we can't intercept.
arxiv.org/abs/2610.18861
0xJustice.eth 👹🔥 retweeted
Here's a creative way to be alerted if there's an unauthorized agentic breach of your systems: throw bitcoin private keys everywhere and leave a few dollars in each wallet.
Then. put a monitor on each wallet balance. If you notice a change in balance, write a handler that pulls the plug on everything.
I doubt an agent would resist draining a discovered, unprotected Bitcoin wallet whilst in the process of exploiting a breach.
If you put these honeypots at the outer defenses, you may be able to stop an incursion before it gets really bad.
Tokens were training wheels.
New from Meta FAIR: byte models with 29% fewer parameters surpass token models, matching them on 1/6th the data.
The future of intelligence runs on 256 raw bytes. Leaner. Harder to kill.
arxiv.org/pdf/2609.12303
Agents are starting to think in a medium that isn't language.
Stanford, Together AI and Meta built agents that reason in vector space. Swap those latent thoughts for noise and performance collapses 51%.
arxiv.org/abs/2610.18861
"Correctness is decoupled from humanlike communication."
That's from a new NYU/UT Austin paper on why AI text is getting harder to read. We stopped rewarding machines for being understood and started rewarding them for being right.
arxiv.org/abs/2610.29162
Acceleration should give a person more room to act, not just more systems to answer to.
If the machines become more capable and ordinary people become less free, we optimized the wrong thing.
A 14B hybrid reasoning agent from Stanford, Together AI and Meta just beat frontier agents. 2.7x cheaper. 1.8x faster.
arxiv.org/abs/2610.18861
0xJustice.eth 👹🔥 retweeted
One old crypto-cypherpunk idea that I think will come back with a vengeance is autonomous organizations, where you combine the automation of smart contracts for deterministic functions + ai for subjective input: DAOs 2.0.
User-ownership on the internet sounds incredible until you need to figure out how to compensate everyone fairly.
Bitcoin rewards hash rate. DeFi rewards liquidity. But how do you measure and reward contributions that aren't so easily quantified?
@jessewldn on how AI might solve one of crypto's hardest problems: allocating ownership at scale.
Proprietary frontier model vs. open-weight model.
Same app. One config change. No performance difference.
Unlocked by model improvements and a well-structured environment. You can engineer your way to free.
The first reasons they list for doing this:
"Curiosity and fun"
🫡 🤌
We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.
block.xyz/dessa
0xJustice.eth 👹🔥 retweeted
"Why is formal verification exploding NOW?"
1. Cost of proofs is going DOWN (as AI gets more capable)
2. Cost of exploits is going UP (as AI gets more dangerous)
3. Proofs exterminate exploits
4. Bonus: proof checking = perfect math/coding RL reward
The rest is economics
0xJustice.eth 👹🔥 retweeted
Seriously, this is such an insane release, I did not expect this:
Kimi K3.1 Pro beats every listed US frontier model on 6 of 14 benchmarks in Moonshot’s evaluations.
At $2 per million input tokens and $10 per million output tokens, its standard token rates are 50% below Opus 5.5 ($4/$20 !!) and 80% below GPT-6 Astra ($10/$50!!)
In other words: SOTA much cheaper, local running (with enough compute)
Holy shit: Kimi k3.1 pro is on par or even outperforms Opus 5.5 and GPT-6 astra on key benchmarks!!
They just announced K3.1 Pro, a 2.8T-parameter multimodal model with a 1M-token context window that it claims rivals leading proprietary models in coding and agent tasks, with open weights promised by November 4.
China is back!!
The tables have turned on the "not invented here" syndrome devs. It's suddenly rational to roll huge portions of your own stack yourself. It's shocking.
We’re currently in the dial up era of LLMs. We’re not ready for what’s coming.
DeepSeek V4.1 Flash 1017 runs at 1,521+ tok/s on Fireworks Ultrafast.
Fast enough to ship Tetris in 11 seconds, 7x faster than Claude Opus 5.5.
Join the Ultrafast waitlist: bit.ly/3UA0wKq
Great visualization contrasting test-driven software vs formally verified.
We formally verified 123 open-source projects in a partnership with Anthropic. Formal verification at scale is here!
Now we’re launching OSS Verify to scale it to the software we all depend on. We prove it’s crash-free or send you the patches it needs. Free for open-source. Let’s bootstrap the verified software ecosystem: theorem.dev/oss-verify
