Felix Brandt

Ironically, the biggest obstacle to broader adoption of LLMs seems to be their lack of reliability, a trait that is otherwise characteristic of computers.
Like the worst of us, they just make up facts rather than admit they don’t know. I appreciate anyone who admits they don’t know something.

Borda strikes back.

Very impressed by ChatGPT4o's new and improved image generation capabilities.

Trump Declares A Trade War On Uninhabited Islands, US Military, And Economic Logic There’s a fundamental problem with Donald Trump’s new trade policy: it fails a test that actual 5th graders can pass. I know this because I tried explaining his “Liberation Day&#8…

www.techdirt.com/2025/04/03/t...

Really hard to tell April Fool's jokes from genuine news this year.
US regrets that far-right extremist Marine Le Pen is convicted of fraud. Trump abandons money laundering law. Musk pays millions to Wisconsin voters. Innocent father from Maryland accidentally deported to El Salvador mega prison.

Happy to see that people in machine learning (ML) are drawing on results from social choice theory. Especially maximal lotteries (ML) are getting some attention because of their robustness against clones. I recently compiled a topical bibliography of my ML papers: www.cs.cit.tum.de/en/dss/brand...

Social Choice and Welfare | Volume 64, issue 1 Volume 64, issue 1 articles listing for Social Choice and Welfare

A Special Issue of Social Choice Welfare on Fair Public Decision Making: Allocating Budgets, Seats, and Probability, edited by Haris Aziz, Edith Elkind, Jérôme Lang, and myself has just appeared.
link.springer.com/journal/355/...

YouTube video by The New York Times Day of Rage: How Trump Supporters Took the U.S. Capitol | Visual Investigations

youtu.be/jWJVMoe7OY0
The fact that all these people have been released from prison epitomizes the absurdity of our times.
It was the largest criminal investigation in U.S. history, with some individuals receiving sentences of up to 22 years.

Fact check: Trump makes false claims about January 6, Europe, NATO and Canada | CNN Politics President-elect Donald Trump made numerous false claims during a wide-ranging Tuesday news conference in Florida, many of them related to foreign affairs and international trade. Here is a fact check ...

Fact-checkers will be very busy for the next four years.

edition.cnn.com/2025/01/07/p...

In other news, Trump announced he could forcefully turn Canada into the US’ 51st state, doesn't rule out military action to take control of Panama and Greenland, NATO partners need to pay 5% (rather than 2%), and threatens that "all hell will break out" in Middle East.

"Donald Trump will be suspended from Facebook and Instagram indefinitely and at least until the end of his time in office" (Mark Zuckerberg, 7 Jan 2021).

"We're going to work with President Trump to push back on governments around the world." (Mark Zuckerberg, 7 Jan 2025).

And you can use LLMs to generate all these translations 😉
Any specific examples you encountered?

Best thing about the machine learning hype: academic spam can be identified much easier. Whenever a suspicious email contains DL, ML, or AI, it can be safely deleted.

During ski season, there are more Dutch people in the Alps than in the Netherlands.

I think I‘d be more comfortable with microblogging platforms if the character limit was more like 3000 rather than 300 characters.

(Hooray! At least I managed to get across this point within a single message.)

Meanwhile, tech billionaires from Amazon, Apple, Google, Meta, and OpenAI have traveled to the president-elect’s residence in Florida and promised to support his inauguration with $1 million donations.

Happy New Year!

AfD is an uprising far-right extremist party that is under surveillance by German intelligence for being anti-constitutional and tied to Nazi organizations. So far, all German parties—across the entire political spectrum—have ruled out forming coalitions with AfD.

He also interferes with international politics, calls the German chancellor an “incompetent fool” who “should resign immediately” and vulgarly insults the former female German chancellor. He writes on his social media platform that “only AfD can save Germany”.

after he was defeated in the previous presidential election). The billionaire, who is not an elected official, now meets with world leaders and congressmen who say it feels like he is the ‚prime minister‘.

becomes a personal advisor of the president-elect (who btw was impeached twice, convicted of felonies, granted immunity from prosecution by supreme court judges that he appointed himself, and vowed to pardon hundreds of violent rioters who stormed the US capitol

Moreover, every 24 hours, he pays $1 million to registered voters of swing states who sign a petition. He is then appointed to lead a newly formed Department of Government Efficiency (which aims at cutting funding for government agencies that regulate his businesses) and

Nobody would believe this story if it were fictional: The (literally) richest man in the world buys the largest social media platform in the world and uses it to sway the presidential election of the largest democracy by systematically spreading falsehoods.

One year ago, I uploaded the videos from my Computational Social Choice class to YouTube (almost 30 hours). Most of it is non-computational “classic” social choice with a special focus on Condorcet extensions.
t.co/O3KCZQTcLC

The limits of p are known. When m→ ∞, p goes to 0 for any n. When m=5 and n→ ∞, the limit is:

They conclude that "the more general case will be extremely difficult to resolve" and may require "entirely new techniques that we cannot now foresee." Kelly (1988) writes that this "is one of the oldest unsolved problems in social choice theory."

Even though intuition and numerical computations suggest that both conjectures are true, a proof remains elusive. The only noteworthy progress was made by Fishburn, Gehrlein, and Maskin (1979), who proved the conjectures for the special cases when either m=3 or n=3.

Let p be the proportion of preference profiles that admit a Condorcet winner. If you fix the number of voters n and increase the number of alternatives m, one would expect that p decreases. Similarly, when fixing m and increasing n (by two). Both statements have been open since 1974!

And that was in 1984. No word-processing. His coauthors say he never used a computer. Instead, he composed entire papers in his mind and wrote them out by hand. His secretary would then type them up.

I coded a little 3-player Scratch game to get my kids interested in programming. Didn't work out, but they love the game:
t.co/oW9NBsqOwj

I have never seen a poor paper by Fishburn.