applied ML @Hudl | pan sports fan | fmr. @CERN ∈ @PhysicsIllinois (#AtlasTTHbb), @BrookhavenLab (#PhenixRPCs) ∈ @ACUedu | takes are all mine
Texas, USA
Joined January 2010
- Tweets1.3K
- Following656
- Followers272
- Likes6.4K
Pinned Tweet
First publication Post-PHD! bit.ly/2l9zN5I #feelsgreatman
Huge hat-tips to the rest of the @HudlAnalysis team! #SSAC17
Austin Basye retweeted
Mach-10 hypersonic reentry of an Orion-like space capsule, simulated in full 3D with native GPU acceleration.
Around 20 seconds to build (and formally verify) the solver, and around 10 minutes to build, optimize, run, visualize, and analyze the simulation on a single H200.
Austin Basye retweeted
At @TerraformIndies we're often asked how, as a tiny start up in California, did we manage to build the cheapest utility scale solar array in North America? I asked @buildmattw to write a post on how he did it.
terraformindustries.wordpres…
Austin Basye retweeted
I see high importance in having a human who is genuinely passionate about a topic, who got obsessed with it, and worked with their own hands to craft an explanation around it, and who tries to share that enthusiasm. Probably all the more important with more AI-generated content.
Austin Basye retweeted
Neuralink president DJ Seo says Elon’s aggressive schedules are rigorous first-principles engineering, and an all-green-light schedule shows 80 to 90% of delays are just assumed
“What I’ve learned is that it’s not really some arbitrary management tactics, but it’s really used as a rigorous engineering and first principle, you know. Again, this is more applicable in the world of atoms, but we have this concept called all green light schedule, which is kind of forcing us to think about if every single light is green, there’s no administrative burdens or delays or no shipping delays or risk or no legacy bottlenecks, how fast can you build it?”
“And when you think from that point of view and vantage point and literally strip away any man-made bottlenecks and you ask a question of how fast can you actually manufacture a chip, how fast can you design a robot and test it. It turns out like, I don’t know what the right number is, but at least 80-90% of it is things that you think are the way it is, just because you don’t really think about it from that perspective.”
Elon Musk says he sets the most aggressive deadline with a 50% probability of being achieved, knowing it will be late half the time, because a five-year schedule just expands to fill five years
“I’m constantly addressing the limiting factor. So, I mean, on the deadlines front, I generally actually try to aim for a deadline that I at least think is at the 50th percentile."
"So it’s not like an impossible deadline, but it’s the most aggressive deadline I can think of that could be achieved with 50% probability, which means that it’ll be late half the time."
"There is like a law of gases expansion that applies to schedules. Like, if you said we’re going to do something in like five years, which to me is like infinity time, it will expand to fully available schedule and it’ll take five years.”
Austin Basye retweeted
Some reflections on AI in math. I found it helpful to simply list out the various parts of what a mathematician does, why math is valuable, and where the funding comes from. From there, what happens if AI gets more capable?
youtu.be/LKrwhjOI8Nc
Austin Basye retweeted
interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had
with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works
already-clean.goodfire.com/
I don't know enough ST to critique the explicit analogy being made.
Yet, insofar as this observation uses the layperson's idea of ST to illustrate the necessity of constraints, imposed by reality, to focus & direct intelligent systems (emergent or constructed)... I 100% agree.
Austin Basye retweeted
I suggest that the right attitude to take toward the OpenAI math release is to imagine we have discovered a notebook full of results from a deceased mathematician. Our job it to make sense of the notebook.
Austin Basye retweeted
a lot of "prompt engineering" tricks are better understood as attempts to trick the user into thinking + expressing himself more clearly
Austin Basye retweeted
The critical distinction between base LLMs (2024 and earlier) and modern LRMs is not symbolic tool use. It's the switch from a transductive paradigm (intuit the answer to the query) to an inductive paradigm (intuit the program/instructions that produce the answer to the query).
They're trained to be inductive, and they perform test-time induction, i.e. test-time prediction of a NL program / reasoning chain. This unlocks entirely new capabilities -- in particular fluid intelligence. Base LLMs, to this day, have ~0 fluid intelligence. LRMs have substantial levels of fluid intelligence.
The performance of LLMs on ARC 1 (a benchmark from 2019) remains ~10-15% today. Scaling them up by a factor ~100,000x got them from 0% to 10%. Meanwhile LRMs the same size or smaller saturated ARC 1 in 2025.
Austin Basye retweeted
I was using ChatGPT for legal advice and it decided to completely hallucinate some preposterous nonsense about how growing wheat to use on my own farm somehow constitutes interstate commerce
Austin Basye retweeted
Hilbert's Entscheidungsproblem was solved. Hilbert's 6th problem never was.
As a result, we got formal verification for math and software, but not (yet) for physics and natural science. So no RLVR, no recursive AI improvement.
At @lanyon_ai, we're rapidly closing that gap.
Good question. I think part of the reason why there has not been any *major* impact on Physics yet can be boiled down to the following few points:
1. *Actual* Physics is not *that* dependent on math. Most of it is applying equations that have been derived a century plus ago to novel/particular problems.
2. More advanced mathematical Physics is still too hard to solve for AI and/or numerical solutions have been more than capable for decades to handle most *actual* Physics problems.
3. A lot of what passes for “advanced theoretical Physics” is just a form mathematical onanism with absolutely no relation to reality, so even if AI makes some kind of advancement there it will not make an iota of difference.
4. The *real hard* Physics breakthrough will require some deep insight that is not contained within the “convex hull” of current human knowledge, and thus largely invisible to the *current* AI systems.
Just my thoughts on the matter.
A throwaway request:
Can we agree to verbally abbreviate "Subject Matter Expert" eg. SME as: "Es-em-EE" as opposed to "Sm-EE" - as the latter consistently maps to the dutiful Mr. Smee of Peter Pan (in my brain specifically) and consistently results in me dropping the thread.
Austin Basye retweeted
“Teodoro. Dig. This ain’t a courtroom. This ain’t D.C. Exhale all that vit-ri-ol, k? It’s a poisonous snake that wants to flood your blood with bogus vibes, man. Nah nah. You take that snake, turn it into a pair of Lucchese’s, take a lil sip-sip of some agave from our madre and just chilllll.
See, Ted, everybody’s always filibusterin’ somethin’. Senate. Mind. Soul. Sometimes you gotta invoke cloture on your own bullshit, brother.
Now look out there. Eleven men. Hundred yards. No amendments. No subpoenas. Just space, time and consequence.
You know why they call it the line of scrimmage, Ted? Neither do I. But it runs down the heart of every man. Which side you on?”
Austin Basye retweeted
I hope the proposals by frontier labs to pace AI research stem from a genuine concern for safety and a recognition of the potential risks posed by future models, rather than a strategic effort to consolidate power and permanently solidify the market dominance of a few top players.
If these efforts are genuine, then oversight must take a more democratic and accountable form, with both national and international components, similar to the Nuclear Regulatory Commission in the US and the International Atomic Energy Agency internationally.
If there is only one organization responsible for safety monitoring, and it happens to be staffed by the same people as the frontier labs, and it is perfectly aligned with them both by incentives and by ideology, then it would be indistinguishable from letting frontier labs self-regulate and self-certify their own safety standards.
Two potential warning signs of regulatory capture to watch out for:
1. Calls to ban open-source AI. Open-source development currently serves as the only real counterweight to the dominance of frontier labs.
2. Attempts to hinder non-frontier research. Models that are one or two generations behind the frontier -- whose level of capability has already been deployed at scale -- are empirically known not to pose safety risks.
As long as we don't see 1 & 2, and we see real international coordination efforts, then I will optimistically believe that the proposals are benevolent.
Austin Basye retweeted
Annnnd now you can multi sort in the grid as well!
Cmd/Cntrl + Click on a column header to enable multi sort so that you can do things like sort by Down then Distance then Play Type
Austin Basye retweeted
You may laugh, but my car has never been so precisely docked in my garage.
This object Orion is looking at is an International Docking System Standard (IDSS) Centerline Docking Target (CDT), Rev E. It's not just cool looking, there is a lot of information embedded in the geometry that a pilot can use. Print your own here!
printables.com/model/1430555…
