Dimitris Vardoulakis (@dimvar) header
Dimitris Vardoulakis (@dimvar) profile picture

ML compilers @nvidia. Views my own. Prev TPU compiler, Closure Compiler @Google; DoctorJS @mozilla; CS PhD @neu_prl; BEng @ecentua. Also on bluesky, mathstodon

SF Bay Area
Joined June 2009
Photo by Dimitris Vardoulakis (@dimvar)Photo by Dimitris Vardoulakis (@dimvar)Photo by Dimitris Vardoulakis (@dimvar)Photo by Dimitris Vardoulakis (@dimvar)Photo by Dimitris Vardoulakis (@dimvar)
Dimitris Vardoulakis retweeted
This is so wrong The reason “Trump didn’t steal the elections” is because we & the pro-voter coalition stopped him EIGHT times over the past 21 months Here’s how: a thread (1/x)
213
2,600
105
9,960
300,433
I'm predicting that this won't happen.
All code will be formally verified within two years.
2
5
386
To clarify, I'm super excited about the AI powered formal verification direction, and I think that owners of projects who want to verify them can make great progress in the next two years. I just think that many software developers won't bother with verification.
1
35
No it hasn't. This is a terrible take. Just as a single example, Trump got elected with ~3M fewer votes than Clinton in 2016.
It seems to me that democracy has on the whole worked very well in America over the last decade, and perhaps better than in any other major country in the West. National outcomes have repeatedly been responsive to underlying voter preferences on meaningful questions, and significant changes in policy have ensued. I’ve been reading Democracy in America this past week, and it’s striking how much it still rhymes nearly two centuries later.
87
Dimitris Vardoulakis retweeted
i've reproduced the results from the MIT blog post today on weight-space programming. but one additional result i have to add: this is a new, step-change emerging capability that's appeared in just the latest set of models.
218
803
268
7,815
600,013
Dimitris Vardoulakis retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
2,363
7,275
5,728
55,918
9,863,961
Dimitris Vardoulakis retweeted
In case you missed it, integer factorization has been proven subexponential (deterministically, worst-case bound)! The hard work comes from Claude's version of the ETH refutation but it seems a little more had to be done to get this result for integer factorization.
22
102
13
942
57,979
Dimitris Vardoulakis retweeted
Can formal verification fully replace fuzzing? Genuine question.
4
31
2
154
10,790
The recent verification work by Anthropic is great. Just one thing I'd like to add to the recent online discussions. Quoting: "we need a strategy for verifying the enormous body of existing software, starting with the most critical dependencies. As we’ve shown, this can be achieved at scale by using generic specs and working at the binary level, sidestepping the need to rewrite entire codebases in another language." This is good as a stepping stone, but not as an end goal IMO. Long term, it'd be better to rewrite the code in memory safe languages built with verification in mind (invariants about the program expressed in the program source). The first verification of a program can be costly, and the delta to re-verify it after each new patch should be small. If we're going to spend all this effort, why keep carrying around C++? anthropic.com/research/forma…
4
2
16
1,015
"Because like ALL things, if the punishment is a fine, it isn’t really illegal for rich people." This is unfortunately very true.
The thing about Google clearing a forest in Finland without permits is that they decided that whatever fine they might face is worth it. It doesn’t cost them as much as not getting the land cleared. Because like ALL things, if the punishment is a fine, it isn’t really illegal for rich people.
1
106
Dimitris Vardoulakis retweeted
Corrupt autocracies prefer to do business with other corrupt autocracies because the profits are personal on both sides, not for the state or the country. No oversight, no investigations.
The US is willing to betray its closest allies in Europe, a market of 450 million people and an economy worth nearly $20 trillion, just to cut shady, corrupt deals with Putin's Russia, an economy roughly the size of Italy's. What an absolutely insane foreign policy.
61
1,761
29
5,897
68,792
Back in 2011, many Samsung phones came with spyware pre-installed, and I had to root my phone and install CyanogenMod to get rid of it. Not surprised at all that they are still doing shady stuff like this. cnet.com/news/privacy/androi…
this account is decompiling a Samsung smart tv and it's actually worse than I was expecting, and I was expecting the worst
1
3
328
Dimitris Vardoulakis retweeted
SCOOP: The State Dept has started asking career diplomats up for promotion if they voted for Trump in 16, 20, 24. Federal law prohibits politically based hiring for career staff. State tells @washingtonsun it's screening for "fidelity" and execution of the president's vision👇
196
2,227
280
6,684
1,062,921
The funk brothers documentary is well worth watching. youtu.be/BL9fF4Jtovc?is=UnjW…
James Jamerson's bass on "What's Going On", visualized note by note. Legend says Marvin tracked him down at a bar that night, and he played it lying flat on the studio floor. One finger, his '62 Fender Precision Bass, and the rest was history.
112
Dimitris Vardoulakis retweeted
The GRU hacked into Jared Kushner's iPhone. Dutch military intelligence (MIVD) then recruited a GRU spy who gave them access to the hacked phone. What MIVD discovered was that Kushner had access to two bank accounts --one Azerbaijani, the other Israeli --where Saudi Crown Prince Mohammed bin Salman and Emirati ruler Mohammed bin Zayed deposited money. It sounds like a slightly far-fetched episode of "Slow Horses," but it's real. And it proves just how eye-poppingly corrupt this administration was and continues to be as its "special envoy" decides the fate of tens of millions in Europe in the Middle East. Simply incredible reporting by Matthew Cole, drawn from his new book on Kushner. vanityfair.com/story/jared-k…
1,449
24,384
1,982
77,605
6,569,012
Dimitris Vardoulakis retweeted
you're just going about your day and you now get to read stuff like "Trump says Sundar Pichai talked him out of renaming the Atlantic Ocean" and you just have to nod and move on with your day. “Oceans are different, sir.” lmao theverge.com/policy/1016751/…
10
165
7
1,008
60,498
Dimitris Vardoulakis retweeted
Replying to @maxious
Very strange how this keeps on happening
9
38
21
623
131,424
Dimitris Vardoulakis retweeted
JD - I've got 15 free minutes before my son's hockey practice ends, so let me see how many of Trump's enrichment schemes you didn't see that I can remind you of. 1/ First one is easy: Trump has executed over 30,000 stock trades, many on inside information, netting him millions.
JD Vance: "I've not seen a single shred of evidence that Donald Trump participates or sanctions in that corruption ... I've not once seen evidence that Donald Trump has enriched himself"
254
2,935
118
11,668
568,605
Prediction: Anthropic will hire junior employees after 2028.
This quoted post is unavailable.
1
29