HNHacker News
TopNewBestAskShowJobs

dgacmu

8,038 karma · joined February 12, 2013

CTO and founder, Enriched Ag. Computer science professor at Carnegie Mellon University. Ex- university of Utah, MIT, Google Brain.

http://www.cs.cmu.edu/~dga/

Of the firm belief that distributed systems is the coolest area ever, followed by computer science in general. Yes, I'm a bit biased.

submissionscomments
dgacmu··on DeepSeek: DeepSeek v4.1 Flash 1017
I decided to try something a little weird research wise the other day analyzing about a thousand older research papers and burned through my $100 chatgpt sub + 3 resets in three hours and only finished half of them. It's the first time I've done that, though.
dgacmu··on Waymo in Toronto
They've been mapping, or whatever they do, Pittsburgh for almost a year now, and it was pretty funny watching the cars navigate around massive snow piles last winter. Curious to see how the rollout goes here. <Insert popcorn emoji.>
dgacmu··on BIPM decides to stop adding leap seconds to UTC
For thousands of years, we used time synchronized to the sun to some approximation that was off by minutes, not microseconds, and nobody cared if the current time relative to the sun was off by a second because they had no way of actually determining that. And then some jerk came along and invented computers and precision timekeeping and suddenly things depended on things like time not appearing to move backwards by as much as a second. :)
dgacmu··on The United Kingdom generates less electricity per person than in the 1970s
But that's a very modern perspective:

1950: Lighting was 30% of total electrical demand [1] 1993-1997: 112 TWh, 9% of demand [2] 2020: 81 TWh, 6% of demand [3]

Couldn't find total used in 1950 but the point remains: Lighting used to be a major source of consumption and now it's pretty minor, and at least the trends in the last 30 years have been driven by efficiency gains, particularly following the energy crisis in the 70s.

[1] https://www.researchgate.net/publication/222531686_Exergy_po...

[2] https://eta-publications.lbl.gov/sites/default/files/lbnl-40...

[3] https://www.eia.gov/tools/faqs/faq.php?id=99&t=3

dgacmu··on Gemini 4 Argon
As of now Argon is available in the android app, at least.
dgacmu··on Gemini 4 Argon
I'm not the person to whom you're replying, but I just tried one of mine:

"I'm interested in understanding the rate of VOC dissipation from fresh oil paint on a surface versus the ventilation rate in the room in ACH. I would prefer solid academic studies of this, followed by analytical evaluation, followed by references to non-quantitative sources. Tell me what's known about this."

that I had run in ChatGPT earlier this week. (Yes, we just had our bedroom repainted, and it's TERRIBLE and I hate it, why do you ask? :-).

I didn't try it first in Gemini Flash, but I did try it in whatever model is used for Gemini Live and the answers were terrible. The ChatGPT answer (I think on medium) was pretty good but it threw in extraneous formulae that weren't actually germane to the analysis. The Argon 4 query I just ran was honestly the best of the answers I've gotten -- it was clear and really well-structured from the perspective of "here are the three big phases that researchers break down paint drying into and what the differing effects of ACH are on those phases".

I'm totally going to give this one a try, after basically giving up on 3.8 flash for most things despite having a free $20 pro subscription (from buying a phone).

Edited to add:

Argon4 transcript: https://share.gemini.google/18Gr5liH1YWR

ChatGPT transcript: https://chatgpt.com/share/6acedc72-ff18-83ea-8909-104cdbc6ac...

dgacmu··on A look at the math that OpenAI claims computes faster Fourier transforms
No, not in this case. The assumption of unit cost arithmetic is simply not physically realizable. It's probably easier to understand the integer case: numbers take bits to store and represent. In a typical computer, you would store numbers as either 32 or 64 bits, where 32 bits could store a number up to about 4 billion (or a signed number between -2 and +2 billion). If you want to store and multiply numbers larger than that, they take up more space and they require more work on the part of the processor, and fundamentally, more energy - adding two 128 bit numbers takes twice the work as adding two 64 bit numbers.

The same applies to the precision of floating point numbers. What Ben is arguing is happening in this case is the AI has found a way to hide work by moving it into operations that, in one (unrealistic) model of computation, are assumed to cost the same regardless of how large or precise the numbers are. Other models of computation have a more accurate model of the cost of these things, but that was not the model the AI chose to apply in this case.

dgacmu··on Teams in Vienna and Beijing have built the first thorium nuclear clocks
I would in general agree with you, but having read through the journal paper: I think the fraction of general audience newspaper readers who could actually read that paper is very low. I come from a science background, spent time dabbling in time and frequency measurement enough to understand some of the concepts, and still found it a painful slog. (I am very much not a physicist!)

I still think it's the right practice for newspapers to always link to the primary source, but in this particular case I understand why they might make an editorial decision not to.

dgacmu··on Teams in Vienna and Beijing have built the first thorium nuclear clocks
The actual journal article, since NYT didn't bother linking to it: https://www.nature.com/articles/s41586-026-11084-4
dgacmu··on Navier–Stokes Lost in Translation
and clearly the computer science perspective is: get rid of the humans and express everything directly in lean so the computers can keep getting work done!

;)

dgacmu··on Subquadratic 3SUM and Subcubic APSP
Yeah, I can't get behind this. By showing one problem had an improved lower bound, they showed that another five algorithms could also be improved, and, yes, they refuted the 3SUM hypothesis and made a lot of conjectures about the hardness of some problems less certain, but now ... now we have to go figure those out more precisely. Which is great. It's progress! And all of the work put into those reductions was what made this one result topple five algorithms, so it's not like it's been wasted work.
dgacmu··on Sharing AI progress in mathematics
I find the claimed matrix multiply result (w<= 2.25) shocking. I hope it holds up.
dgacmu··on Subquadratic 3SUM and Subcubic APSP
Absolutely! This is unlikely to directly, or even in the next 20 years, result in anything practical. But it has a very similar feel to Stothers' and then Virginia Williams' earlier improvement on matrix multiply, where his thesis and her first paper were followed by a dozen others finding ways to build on it after 20 years of seeing no progress on the problem at all. None of them have resulted in anything practical but who cares, really? Understanding the problem better is good and maybe some time in the next hundred years it'll result in an improvement in practice also. Or not. :)
dgacmu··on Subquadratic 3SUM and Subcubic APSP
It feels different to me from the CS side - this paper in particular feels likely to open up new research instead of closing it off, and I find that really exciting and a worthwhile use of AI. Showing that there's a (completely impractical but who's counting) algorithm better than the previously hypothesized lower bounds seems like the kind of thing that will inspire a scramble to keep beating it (and figure out the true lower bound). I give this one a thumbs up.
dgacmu··on Nano Banana 2.1
https://xxcancel.com/googleaistudio/status/21075013038909155...

for those who don't wish to login.

dgacmu··on Improper redaction reveals Google Data Center water and electricity usage
For most datacenters the actual largest source of water use is indirectly from power generation (coal, nat gas). Solar reduces that a lot also.
dgacmu··on Reasons I didn't become an EMT, ranked
I was certified as an EMT and spent some time volunteering in the ER, and while I did see some things that were somewhat traumatizing when I was young, I don't think I'm worse off because of them. Seeing someone my age die hurt. But it also made me appreciate the fragility of life a little more and turned me into a more careful driver (which was a good thing for 20 or so year old me). That said, it was only a day or two a week for a year or two, so much less exposure than someone full time gets. But I'm really glad I did it.

(But, y'know, this is a very ymmv kind of thing. I enjoyed cadaver lab and thought attending an autopsy was fascinating; other people had very different reactions, etc. no idea what causes us to differ there, but differ we do!)

dgacmu··on Clef: Open-weight decision models, and new RL fine-tuning platform
Fair. Though in this case my actual classifier (dino v3 features fed into a very shallow DNN) works surprisingly well and is insanely faster than using an LLM -- so it really was the one shot performance I was hoping for.
dgacmu··on Clef: Open-weight decision models, and new RL fine-tuning platform
Since it supports vision, I tested it (an 8 bit quant) against one of my pet problems of coin classification (US coins, but not beauty shots of them, just lots of cell phone images on a piece of paper). It did .. horribly. About 41% accuracy. My local gemma4:27b gets 53% and runs 4x faster. Not too surprising but seemed worth poking at it.
dgacmu··on Livenerf: Has Opus 5.5 been nerfed yet?
Except for that one year when they completely swapped the meaning of the Cumulus line, which I think was 2008 with the cumulus 9 to 10 transition. It went from a neutral shoe that was good for people with high arches to more of a stiff stability shoe. The complainers aren't _always_ crazy. :-) (That doesn't mean the cumulus 10 was worse, of course, it just was a more substantial change that affected the type of runner the shoe was designed for.)

Wow the old grumpiness that lingers in my head from losing my favorite shoe. Who knew? Now I'm old and heavier and run in the Nimbus and am happy again. But you're right, of course, that most of the model changes are just fine and people like to complain.

dgacmu··on Ten lines of code that changed my world
Most definitely the juvenile phrases.

I think my second BASIC program was the "guess a number" one -- that seemed really popular in intro basic programming things in the 80s also since it showed keyboard input and conditionals and was, uh, fun-adjacent.

dgacmu··on We're gonna need a lot more mathematicians
We've been those ants for a million years, and only in the last 100 did we start to wrap our heads around quantum physics. We are the purpose behind creating LLMs. There's plenty in the universe we don't understand, and it's very human to keep striving to do so.

If you believe that math is discovering, it's natural to think that all of that AI math already exists and is just waiting for us to find ways to discover and understand it.

Don't write us out quite yet. :)

dgacmu··on An airport cooled by natural ventilation
This is so humidity dependent though, so it's probably more useful to speak in terms of the heat index.

I didn't mind 100F too much when I lived in Salt Lake; it was hot but that's what drinking lots of water was for. In Pittsburgh, 93F and higher make me want to curl into a ball and hug an air conditioner. In the 90s the heat index reaches as high as 110 here. Ick.

dgacmu··on Initial DIY cleanroom experimentation
I don't know how general it is, but my dentist and kid's school kept the HEPA filters after covid. It's at least... a little better?

(Speaking personally, my home indoor air quality is now awesome and I love it. Wish it hadn't taken a pandemic to clue me in, though.)

dgacmu··on Creatine uptake enhances antitumor immunity
Your body can synthesize creatine, so it's not essential, but meat is a source of exogenous creatine and eating it results in higher stored creatine levels, which is slightly good for muscle power endurance in the ~5-15 second effort range. Supplementing produces slightly higher relative improvements in creatine levels in vegetarians and vegans than it does in people who eat meat.
dgacmu··on Tokens too cheap to meter
I've evaluated some of these and in almost all cases they failed to outperform Intel's ridiculously optimized CPU regex library, hyperscan.
dgacmu··on Launch HN: Coverage Cat (YC S22) – Umbrella insurance via your personal agent
seems reasonable to me! Thanks for responding.
dgacmu··on Launch HN: Coverage Cat (YC S22) – Umbrella insurance via your personal agent
Perhaps I'm not making myself clear: There might be reasons I _shouldn't_ be happy with it! Maybe I'm overpaying, maybe I'm missing out on subtleties of one insurer vs another, etc. And I mean, hello, I'm posting here, of _course_ i'd rather not talk to a salesperson on the phone.

Your post actually inspired me to go raise the amount of my coverage. I was paying attention. It triggered some thinking. It even triggered action. But maybe I wasn't paying attention to the part you wished I was.

dgacmu··on Launch HN: Coverage Cat (YC S22) – Umbrella insurance via your personal agent
I'm probably your target audience - I have umbrella insurance - but I just grabbed it through my existing GEICO/travelers insurance and it was basically painless. It isn't so expensive I felt compelled to shop massively for it and nothing I read above convinced me I'm wrong there?
dgacmu··on RSA-896
If you look at the numbers, he managed about 50% utilization of those 2048 GPUs over 10 days, so he was probably sneaking in factoring work between training runs.
← PreviousPage 2 of 34Next →