I'm in crypto to eliminate the middleman. Disintermediate all human coordination. Finance, commerce, social, governance (corporation, even state level), information curation, internet infra, physical infra, content creation, science, property rights, employment...

guess the model

Opus 5.5 is surprisingly pleasant to talk to, it feels much less overbearing than Fable 5.1

realizing more and more that RL is an art not a science because compute is the limiting factor also realizing that its really hard to get agents to drive RL research because they rarely even try to be scientific

its incredibly important that you have this arc and then eventually abandon it

What piece of writing or rhetoric that you have read had the greatest density of emotion per word?

I find myself constantly searching for better software and experiences, much more than I did before AI I know what is possible with someone like @steveruizok and tldraw and I dream that this can exist for any app or website I touch.

We had a special guest at THE BASED SHOW Featuring @mteamisloading [Co Founder and CPO of @spire_labs , @baibai_cx ] Timestamps: 0:53 Welcome Matthew 1:08 Who is Matthew, and why crypto: killing the middleman 2:51 -From MEV searching to Ethereum research 3:47 - Building Spire Labs and synchronous composability 4:15 - The pivot from infra to a consumer app 4:57 - The 3 problems DEXs have: LPs, traders, builders 10:08 What a prop AMM is, and how it differs from Uniswap 14:24 - Does retail actually care about fees? 16:52 - The Walmart playbook: a 2x best-price guarantee 18:14 - Marketing gimmick or real? Claims so far 18:53 - What's next: tokenized stocks and the altcoin price gap 22:18 - Multi-chain roadmap: Robinhood Chain and beyond 23:15 - Built for market makers first 25:10 - Traction: 10K+ users in the first 4 weeks 27:18 - Funding, token plans and community ownership 28:19 - Points Season 1: "How wrecked are you?" 29:32 - Wrap-up

Analyzed a few thousand hours of agentic runs with @typesafeai Jev - turns out it's: • The best option I've tested at measuring progress and estimating completion • Dangerous if you use it for detecting harmful commands (more on that below) and • Not very good at catching models being lazy Actual prompts and results in the article: southbridge.ai/blog/jev-watc…

Also have been playing with @typesafeai Jev, insane! So many immediate use cases and new apps are possible. What a time to be a builder! Sharing some experiments here starting with: Keystroke oracle / predictive launcher: Your launcher ranks by aliases, fuzzy match, and habit. Jev reads intent: type "the pdf I just downloaded" and the newest PDF is already the top hit with a full confidence on every keystroke, in ~100 ms

Momentum is provably conserved in this physics sim! Bend2 is very nice!

Bend 2 is here! It is a new programming language that blocks AI mistakes via *proof checking* - the same technique big AI labs used to solve open math problems, like Navier-Stokes. It is also very fast, and runs on GPUs. Watch the video. Link in the comments.

1/ Fast Computer Use is now solved with @typesafeai Jev + Cua Driver. Available in development preview for macOS, Windows, and Linux. We call it jev-use. Draft #3943: github.com/trycua/cua

Pi is getting rewritten, and @badlogicgames is mildly optimistic!! I posted about the upcoming changes before, but some other points to highlight: - durable tasks: saving what is running, state, dependencies and recovery - shared task scheduling between extensions - flexible document management to manage application state Everything is still back to state machine fundamentals lol. The design choices and changes are up in `pico` branch, take a look as its quite different from the existing pi harness. This will open up a lot more use cases: - different interfaces with pi agent, browser, etc. - durability from recovery across tasks - better state management than just context history in documents - syncing state with "chords" across different envs Idea seems to be a fully customizable harness that can build any production agent, any client interface, as well as better plugin integrations across the board :)

fast computer use delegated by frontier models will be an insane way to get stuff done

Realtime AI like Jev embedded into all UIs for custom visualizations and component views is going to be insane

My dream for Jev is that it can generalize Google's ranking/classification so that I can use it over code, my markdown notes, call transcripts etc. No LLM or vector search or whatever is even close to human intuition yet it only has 32k context limit though 😢

My dream for Jev is that it can generalize Google's ranking/classification so that I can use it over code, my markdown notes, call transcripts etc. No LLM or vector search or whatever is even close to human intuition yet

Glad to see our friends at Blockspace releasing MPBC, which I think is one of the most interesting PBS developments recently. Putting it simply, it provides a residual path for orderflow landing via direct access to the block relay. While it originally targets competing PBS block builders, we're planning to rollout its integration in TOOL, to provide the lowest-latency transaction inclusion for Ethereum users, in cases when there's no direct connection to the block proposer.

onchain aggregator - timely given the games liquidity providers are playing - expensive gas wise but cheap in dollar terms on monad

Trade spoofing via malicious Prop AMMs or Uniswap V4 hooks is no longer a problem. Introducing Moose: the world's first fully onchain aggregator. It is the fastest, most private, and most decentralized aggregator. It has the best execution. Only on Monad.