Dimitris Vardoulakis

3.5K posts

Dimitris Vardoulakis banner
Dimitris Vardoulakis

Dimitris Vardoulakis

@dimvar

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 加入时间 Haziran 2009
928 关注687 粉丝
Dimitris Vardoulakis 已转推
Elizabeth Warren
Elizabeth Warren@SenWarren·
If you paid even a penny in federal income tax last year, you paid more than: Tesla Southwest Disney Live Nation HP United PayPal CVS Health Palantir Citigroup PG&E 3M That's right. They paid $0 in federal income tax. It's time for big corporations to pay their fair share.
English
9.2K
6.7K
22.4K
2.8M
Dimitris Vardoulakis 已转推
Rock'n Roll of All
Rock'n Roll of All@rocknrollofall·
A father playing and making duet on The Beatles' Don't Let Me Down with his child might be the best video you'll watch today.
English
112
1.4K
9.9K
379.4K
Dimitris Vardoulakis 已转推
Steven Tavares
Steven Tavares@eastbaycitizen·
I’ve covered Eric Swawell since he was a member of the Dublin City Council. Shortly after being elected to Congress in 2013, his behavior towards women was known by all levels of our local government and the Alameda County Democratic Party.
English
3.8K
2.4K
12.7K
2.6M
Dimitris Vardoulakis
@NittaSatya @prfsanjeevarora I was at Google at the time when the controversy happened where an engineer claimed lamda was "sentient". It was an ok chatbot, but nowhere near chatgpt, not even close.
English
0
0
0
22
Sanjeev Arora
Sanjeev Arora@prfsanjeevarora·
I don't think Google had a chatGPT equivalent in late 2022. (But I agree that if they would have had one, they would probably not have released it.)
Jared Duker Lichtman@jdlichtman

Without taking a stance on him personally, Sam Altman @sama is the individual most responsible for the current dynamics in the AI industry landscape: It's defining feature is iterative deployment of increasingly capable models, across all frontier labs. Before the ChatGPT moment, Google Deepmind had the strongest models internally, but Demis Hassabis @demishassabis had been reluctant to release to the public (per @scmallaby Mallaby’s recent biography), for fear of societal impacts and race dynamics. History will judge whether iterative deployment will be better for society absorbing technological change gradually, or exposing downside risk.

English
5
0
28
13.2K
Dimitris Vardoulakis 已转推
Easy
Easy@NotSoEasyMoney·
HOW IS NOBODY TALKING ABOUT THIS!?!?!?!? THEY PRINTED 5 BILLION OF THEIR OWN TOKENS THEN WITHDREW IT AS USDC!?!?!?!??!
Easy tweet media
English
988
4.5K
15.2K
3.8M
Dimitris Vardoulakis 已转推
Travis Akers 🇺🇸
Travis Akers 🇺🇸@travisakers·
Preemptively stating that you are going to pardon your entire staff and yourself before leaving office is one of the biggest admissions of guilt I have ever seen.
English
1.7K
8.7K
50.9K
588.1K
Dimitris Vardoulakis 已转推
Melanie D'Arrigo
Melanie D'Arrigo@DarrigoMelanie·
What an odd way to admit that your staff are committing crimes, and to let them know they won’t get in trouble when they commit more.
Melanie D'Arrigo tweet media
English
2.1K
7.6K
15.6K
259.3K
Dimitris Vardoulakis 已转推
Mike Quigley
Mike Quigley@RepMikeQuigley·
Pete Hegseth has cut ALL educational funds for active-duty service members at Harvard and Princeton (despite him attending these schools) in favor of Christian-affiliated colleges. Our service members deserve the support they need to finish their studies, regardless of which school they attend or their religious beliefs. I’m again calling for Hegseth’s resignation on the grounds of limiting academic freedom and disregarding the separation of Church and State.
Fox News Politics@foxnewspolitics

WARRIORS FIRST: Hegseth cuts 93 military fellowships at Harvard, Princeton, replacing them with Hillsdale and Liberty University. New schools were selected for "intellectual freedom" and limited ties to foreign adversaries. Hegseth holds degrees from both Princeton and Harvard's Kennedy School. foxnews.com/politics/hegse…

English
296
3.4K
10.7K
512.9K
Dimitris Vardoulakis 已转推
Mike Young
Mike Young@micyoung75·
Applebaum's piece is worth reading slowly because the specific details are doing work that the summary can't. Danish military commanders - inside a NATO alliance the United States founded - had to sit in a room and war-game whether their forces would shoot down American planes and kill American soldiers. Some of them still haven't fully recovered from running that exercise. The most popular app in Denmark during Applebaum's visit was one that identifies American products so users know not to buy them. NATO has invoked Article 5 exactly once in its history. On behalf of the United States. After September 11th. Allied troops went to Afghanistan and some of them died there. Trump told reporters those allies "stayed a little back, a little off the front lines." The families of soldiers who didn't come back heard that. Now Trump is in the middle of a war in the Persian Gulf with the Strait of Hormuz locked, oil prices spiking, and he's telling NATO allies - the same ones he insulted and tariffed and threatened - that he's "demanding" they come help solve a problem his own decisions helped create. Applebaum's conclusion is precise: he doesn't connect what he does on one day to what happens weeks later. Allied leaders have drawn their conclusions. The rupture, as Mark Carney called it, isn't coming. It already happened.
Mike Young tweet media
Anne Applebaum@anneapplebaum

Trump has insulted and tariffed his European allies, persuaded Denmark to prepare for a US invasion and, by pressuring Ukraine and not Russia, encouraged Putin to keep fighting. All of which he has forgotten. theatlantic.com/ideas/2026/03/…

English
203
4.3K
14.2K
1.2M
Dimitris Vardoulakis 已转推
Lorraine Evanoff
Lorraine Evanoff@LorraineEvanoff·
Fucking outrageous. And highly illegal. "Trump has grifted his entire life. Now he’s just taking it. The State Department transferred $1.25 billion in foreign aid to Trump’s Board of Peace, pulling $1 billion from international disaster assistance, $200 million from peacekeeping operations, and $50 million from international organizations. Money that Congress authorized for hurricanes and refugees, moved without a congressional vote, into a fund that Trump created by executive order and controls personally. When reporters asked the State Department about it, a spokesperson said they had nothing to announce at this time. The Board of Peace has one defining characteristic. Trump controls it forever. He named himself chairman for life. No audits. No transparency requirements. No conflict of interest rules. Countries pay $1 billion into a fund he runs to get a seat at the table. It has transferred nothing to Gaza, disclosed nothing about its spending, and received $1.25 billion of your disaster relief money without a word of explanation. When he leaves the White House he keeps the fund. That is not a loophole. That is the design." He’s Not Grifting Anymore. He’s Just Taking It. open.substack.com/pub/meidastouc…
English
348
16K
29.1K
1M
Dimitris Vardoulakis 已转推
Jon Favreau
Jon Favreau@jonfavs·
Our president is a deranged lunatic and should clearly be removed from office. He’s only there because of Republican Members of Congress.
Jon Favreau tweet media
English
4.2K
6.8K
37.9K
664.9K
Dimitris Vardoulakis 已转推
Bernie Sanders
Bernie Sanders@BernieSanders·
One month after starting the war in Iran, this is the statement of the President of the United States on Easter Sunday. These are the ravings of a dangerous and mentally unbalanced individual. Congress has got to act NOW. End this war.
Bernie Sanders tweet media
English
10.8K
50.6K
240.1K
5.9M
Dimitris Vardoulakis 已转推
Melanie D'Arrigo
Melanie D'Arrigo@DarrigoMelanie·
Trump’s kids buy into drone companies Trump cancels existing drone contracts Trump’s kids’ companies get military contracts Trump starts wars Trump’s kids’ try to sell their drones to the countries being attacked because of Trump’s wars 👉🏻 This is what corruption looks like.
Melanie D'Arrigo tweet media
English
825
20.7K
36.6K
731.2K
Dimitris Vardoulakis 已转推
Adam Schiff
Adam Schiff@SenAdamSchiff·
The President starts a war with Iran. Now, his family looks to profit from it. The corruption is enough to make you sick.
Adam Schiff tweet media
English
2.2K
5.1K
14.9K
206K
Dimitris Vardoulakis 已转推
Mike Levin
Mike Levin@MikeLevin·
The Army Chief of Staff, a combat veteran with tours in Iraq and Afghanistan, asked Pete Hegseth why he was blocking the promotions of two Black officers and two female officers who had earned them. Hegseth refused to answer. Then he fired him. Nine U.S. officials told NBC News that Hegseth has blocked or delayed promotions for more than a dozen Black and female senior officers across all four branches of the military. Hegseth has now fired or sidelined more than a dozen generals and admirals. He is an out-of-control, unqualified former TV host and nobody in the Republican Party will say a word about it because they don’t want to make Trump angry. nbcnews.com/politics/natio…
English
3.4K
26.1K
67.6K
2.8M
Andrej Karpathy
Andrej Karpathy@karpathy·
LLM Knowledge Bases Something I'm finding very useful recently: using LLMs to build personal knowledge bases for various topics of research interest. In this way, a large fraction of my recent token throughput is going less into manipulating code, and more into manipulating knowledge (stored as markdown and images). The latest LLMs are quite good at it. So: Data ingest: I index source documents (articles, papers, repos, datasets, images, etc.) into a raw/ directory, then I use an LLM to incrementally "compile" a wiki, which is just a collection of .md files in a directory structure. The wiki includes summaries of all the data in raw/, backlinks, and then it categorizes data into concepts, writes articles for them, and links them all. To convert web articles into .md files I like to use the Obsidian Web Clipper extension, and then I also use a hotkey to download all the related images to local so that my LLM can easily reference them. IDE: I use Obsidian as the IDE "frontend" where I can view the raw data, the the compiled wiki, and the derived visualizations. Important to note that the LLM writes and maintains all of the data of the wiki, I rarely touch it directly. I've played with a few Obsidian plugins to render and view data in other ways (e.g. Marp for slides). Q&A: Where things get interesting is that once your wiki is big enough (e.g. mine on some recent research is ~100 articles and ~400K words), you can ask your LLM agent all kinds of complex questions against the wiki, and it will go off, research the answers, etc. I thought I had to reach for fancy RAG, but the LLM has been pretty good about auto-maintaining index files and brief summaries of all the documents and it reads all the important related data fairly easily at this ~small scale. Output: Instead of getting answers in text/terminal, I like to have it render markdown files for me, or slide shows (Marp format), or matplotlib images, all of which I then view again in Obsidian. You can imagine many other visual output formats depending on the query. Often, I end up "filing" the outputs back into the wiki to enhance it for further queries. So my own explorations and queries always "add up" in the knowledge base. Linting: I've run some LLM "health checks" over the wiki to e.g. find inconsistent data, impute missing data (with web searchers), find interesting connections for new article candidates, etc., to incrementally clean up the wiki and enhance its overall data integrity. The LLMs are quite good at suggesting further questions to ask and look into. Extra tools: I find myself developing additional tools to process the data, e.g. I vibe coded a small and naive search engine over the wiki, which I both use directly (in a web ui), but more often I want to hand it off to an LLM via CLI as a tool for larger queries. Further explorations: As the repo grows, the natural desire is to also think about synthetic data generation + finetuning to have your LLM "know" the data in its weights instead of just context windows. TLDR: raw data from a given number of sources is collected, then compiled by an LLM into a .md wiki, then operated on by various CLIs by the LLM to do Q&A and to incrementally enhance the wiki, and all of it viewable in Obsidian. You rarely ever write or edit the wiki manually, it's the domain of the LLM. I think there is room here for an incredible new product instead of a hacky collection of scripts.
English
2.8K
6.7K
56K
19.9M