Art Scott

Amazingly, according to Artificial Analysis, good old Ox Alpha (GLM-5.3 Flash), which I'm currently deploying in ArcturusChess.com, remains on the Pareto frontier for AAII score vs. cost-per-task. Not sure what thinking level that is, though. In the web app, Max sometimes blows out the turn clock. Astra's updating the web app now to allow the user to switch between High and Max thinking levels on the fly.

Some major scientific benchmarks are badly broken, and frontier models actually do much better on them than the official scores show.

In this first version, things are somewhat idealized -- fluxons travel perfectly ballistically, and (reversible) interactions are perfectly elastic. The simulation is (mostly) deterministic and 100% reliable, but there are a few (intended) failure modes which you may encounter!

The first simple version of this project -- an (educational) puzzle game that lets people learn and play around with the BARCS (Ballistic Asynchronous Reversible Computing in Superconductors) model of computation -- is now live at fluxonaut.netlify.app/. Give it a try!😃

Just give me some gigawatts, and I'll token-maxx ya using my ultra-efficient adiabatic reversible computing (ARC) hardware. 😁

I started programming around age 11, first at summer camps and then using a little TRS-80 Color Computer I had bought for myself using saved-up birthday and Christmas money. But I got bored with coding by midway through grad school — it just wasn’t even challenging enough any more, and my friends and I had already won the world programming championships as undergrads back in ‘91 (ACM ICPC). Grinding away as a software engineer at some faceless mega-corporation held zero appeal for me. That (together with a repetitive stress injury, thanks to Emacs and an old mushy Sparcstation keyboard) helped motivate me to focus more intensively on my research career and finish my PhD. If the field you spent the last 15 years becoming a top expert in no longer even feels challenging for you, then just try something harder, amirite? That’s when I switched fields from CS/AI to the physics of computation, and began my 30-year journey into reversible computing research, which has been an endless source of challenges. Perfecting this technology is a never-ending quest; in reality, of course you can approach but never quite get to 100% efficiency. And even just doing well enough at it to crush SOTA conventional CMOS digital tech is already a magnificently difficult problem, as von Neumann might say. But my point is, for those CS people and programmers who are going through an existential crisis right now because AI has made the focus of all of their previous cognitive effort seem so downright easy that it’s no longer interesting — welcome to where I was 30 years ago. The solution, of course, is to do what I did — TRY SOMETHING HARDER. Programming is too easy, but you now have a certifiable genius-level AI at your fingertips, what can you do with that? What grand civilizational challenges are you now suddenly equipped to tackle? Instead of just moping around 😩 because you can no longer pay the rent by being a code monkey, THINK BIGGER. Today, no problem is too hard for a motivated, smart person to tackle. Just grab an AI, or a well-orchestrated agent swarm, pick a hard problem in any field, and go out there and conquer the world! 🦾😤✌🏼 AI is a wonderful enabler, and it’s never too late to reinvent yourself. So, go get em! 😁✌🏼

In ecology, one finds that the most stable ecosystems are always the most diverse ones, with many competing species. Civilization works in the same way. Many powers, of many sizes. The notion of one world government was never really going to fly — it’s too prone to becoming corrupted (single points of failure and all that), so whenever one power starts getting too big for its britches, it meets resistance from others. So we have a multipolar world, and that’s probably for the best. One sees this play out in crypto as well; it’s not healthy for the Bitcoin ecosystem for any one mining pool to have a majority of hashing power, so there are always at least a few large players. I think the same will always hold true in AI as well. No one lab is going to outstrip all the others and become a singleton ASI. Everyone can see that this wouldn’t be a good thing for the world. (Absolute power corrupts absolutely, you know.) There will always be smaller labs, open source and so on. So yeah, no one power is going to eat the lightcone; even if the big AI labs collude with each other, they will face resistance to any kind of play for total world dominance. The world does have other significant powers, and tech isn’t the entire pie, not by a long shot.

Haha it's kind of funny how the phrasing here almost makes it sound like I lead all of @SandiaLabs -- to be clear, obviously I don't, and I'm not currently even at Sandia right now, but on entrepreneurial leave at @VaireHQ anyway. 😊

I should try it out soon.

A former student reminds me that in my research group back at UF, we were already writing papers envisioning automated agents everywhere about a quarter-century before AI agents were cool 😎

Yo, @VaireHQ mentioned! 😁🙌 futuristspeaker.com/artifici…