okintheory

"A strict complexity gap between gradient-based and derivative-free Lipschitz convex optimization" has been known for decades. It's covered in standard textbooks like Nesterov's "Introductory lectures on convex optimization". That AI result is about establishing the gap in a fairly niche accuracy regime.

This was clearly written by someone with too little exposure to history and (comparably) too much to academic economics. No one else could think Coase belongs on such a list and forget Orsted/Faraday/Maxwell (initially...). And if you think John Locke did something important beyond adding philosophical veneer to capitalism as it was already practiced, you need to read Meiksins Wood's 'The Origin of Capitalism'.

How could any European govt use MS after Trump ordered MS to sanction an ICC prosecutor and MS complied? I imagine they're all trying to walk away

I don't think that's a great example. If Kahneman claimed not to be susceptible, it would have greatly undermined his claims about the universality of these phenomena: many other people would presumably also not be susceptible.

I settle this question for myself every month: I try asking ChatGPT and Gemini for help, but in my domains it fails miserably at anything that looks new. But, YMMV, that's just the experience of one professional mathematician.

I think you're misunderstanding the point this paper is trying to make. They're interested in trying to distinguish whether AI is capable of solving new math problems or only capable of identifying existing solutions in the literature. Distinguishing these two is difficult, because self-contained math problems that are easy enough for LLMs to address (e.g. minor Erdos-problems) may have been solved already as subcomponents of other work, without this widely known. So when an AI makes progress on such an Erdos problem, we don't know if it had a new idea, or correctly identified an existing but obscure answer. This issue has been dogging the claims of AI solving Erdos problems.

Instead, here you get questions that extremely famous mathematicians (Hairer, Spielman) are telling you (a) are solvable in <5 pages (b) do not have known solutions in the literature. This means that solutions from AI to these problems would perhaps give a clearer signal on what AI is doing, when it works on research math.

You're wrong. The mistake could have been unfixable. That happens quite frequently (see: countless retracted claimed proofs of major results by professional mathematicians).

The thought police already arrived, see Columbia grant cancellations and Mahmoud Khalil [1].

[1] "Khalil is a “threat to the foreign policy and national security interests of the United States,” said the official, noting that this calculation was the driving force behind the arrest. “The allegation here is not that he was breaking the law,” said the official." https://www.thefp.com/p/the-ice-detention-of-a-columbia-stud...

I also can't tell whether Stanford is joking, but the notion that he's a good fit for the job of biology professor is definitely funny!

It's nice to live in a world where actions have consequences. When the media coverage got too much, Marc Tessier-Lavigne finally had to resign as president of Stanford, so he could focus on his job as a Stanford professor.

Interestingly, the asymptotically fastest known algorithm for minimum weight bipartite matching [A] uses an interior point method, which means it's also doing Riemannian optimization in some sense.

[A] https://www.quantamagazine.org/researchers-achieve-absurdly-...

It's right there, in the press release:

>>> Jonathan Friedman, Sy Syms director of PEN America’s U.S. Free Expression programs, said:

“The irony cannot be lost here: government officials have used their positions to muscle out a scholar of authoritarianism from a prestigious lecture," <<<

Ah yes, Minority Report, that story we're so eager to repeat.

But, the starting point of Neural Networks in the ML/AI sense, is cybernetics + Rosenblatt's perceptron, research done mathematicians (who became early computer scientists)

Absolutely. This makes very little sense, IMO, and is a bad look for the committee, trying to claim 'physics' for something that clearly is not.

These problems are literally already solved? Of course, the IMO problem designers make sure the problems have solutions before the use them. That's very different than math research, where it's not known in advance what the answer is, or even that there is good answer.

Interestingly, the same guy also works on making 'theory-only' algorithms work well in practice [1]. But, it seems like that takes another 20 years -- [1] is building on a theory breakthrough from 2004 [2], but these algorithms are only starting to work in practice in 2024, IIUC. I guess that means there's hope for practical min-cost flow algorithms in 2044.

[1] https://arxiv.org/pdf/2303.00709

[2] https://arxiv.org/abs/cs/0310051

I'm not arguing that X/Twitter is doing something that prevents it from being protected by section 230 as it exists. But, when Congress granted these protections, I think they had in mind things like forum moderation to remove hate speech, not selectively censoring journalists that write critically of a company. So one could rethink whether companies that censor journalists should be protected.

When Ken Klippenstein obtained footage of an FSD Tesla causing a major accident [1] and wrote about it, he got blocked from being found through X/Twitter search [2]. I remember verifying the search block on my phone at the time.

If I were a US lawmaker, I'd take a moment to reflect on whether a company such as X/Twitter, which actively censors journalists to protect the financial interests of its owners, should really be granted Section 230 protection.

[1] https://x.com/kenklippenstein/status/1612848872061128704 [2] https://x.com/kenklippenstein/status/1604574747945275396

When was mainstream media not advertising driven? When was media not beholden to corporate interests?

Also, just to add, 'quantumalgorithmszoo.org' is quite out of date: For example, they claim a polynomial speed-up for network flow, but that's based on the SOTA in 2007. These days, there's an almost-linear time classical algorithm [A], and no matching quantum algorithm (and also the idea that you would want to use a quantum computer to shave a factor n^(0.0001) is ridiculous). Now, all these statements are only about asymptotics, but don't get me started on practicalities: The idea that in the next 50 years you would use a quantum computer for any problem with only a polynomial speed-up is silly.

[A] https://www.quantamagazine.org/researchers-achieve-absurdly-...

You should be very skeptical of all those applications except Shor.

HHL: Here's a quote from Ewin Tang [1]: "We know that quantum computers can “efficiently solve” high-dimensional linear algebra problems; however, this assumes that we have some way to evolve a quantum system precisely according to input data, a much harder problem than the linear algebra itself."

[1] https://ewintang.com/blog/2019/01/28/an-overview-of-quantum-...

Grover's search: This is a speed-up from 2^n to 2^sqrt(n). Impressive, but there's not a lot of exp-time algorithms that people ever run. They go for heuristics instead.

Quantum fourier transforms: This is a tool, it's cool, but needs an application. I haven't seen a serious proposal for using it somewhere where a classical algorithm wouldn't do better.

There are lots of problems for which algorithms (with no data/learning required) are trivially superhuman, in the sense that they perform tasks no human could hope to do. AlphaFold is not 'only' an algorithm, as it also needs data to perform well. But this doesn't seem like a satisfactory reason to call AlphaFold "ASI" and not apply this term to your run-of-the-mill mixed-integer-program-solver for some difficult scheduling problem.

I didn't read your comment as a defense, only as an explanation. And as an explanation, you're right: The pressure is getting to way too many people. But how can you fix it? I'm afraid good answers need very deep change. Society lets a few 'winners' (whether by cheating, effort, good luck, or anything else) reap too much of our collective rewards.

Very glad to see this. The Stanford Daily did a great job reporting this. Reposting a comment that I found instructive from the discussion on this piece [1]. About the Genentech report [2] which made MTL look very very bad.

``` APersonWhoCanRead 3 months ago

It seems to me that the linked report goes as close as possible to accusing MTL of fraud as one could hope given that it's coming from Genentech lawyers that are trying to keep the company out of trouble:

"In order to assess whether the 2009 Nature paper contains duplicate images, the diligence team consulted an independent, outside expert who specializes in detecting image manipulations in scientific publications. This expert concluded that two sets of figures, Figures 1d and 5e and Supplementary Figures 9c and 17c, include duplicate images. The expert also concluded that a Western blot panel for Caspase 6 in Supplementary Figure 6d appears to include a composite of two images. We have not determined how these anomalies occurred."

"Genentech scientists and research associates had difficulty reproducing certain results reported in the 2009 Nature paper, in particular, the binding interaction between DR6 and N-APP (the N-terminal portion of APP). Prior to publication of the paper, employees other than the authors performed binding experiments that showed inconsistent results – sometimes binding between DR6 and N-APP was detected, and other times, it was not. Some of the employees who performed those experiments attributed the inconsistent results to variability in the purity and quality of the reagents used." --> Clearly, some employees attributed the inconsistent results differently - I'm guessing as fraud. --> These determinations were made before the paper, which contained fabricated data (c.f. above), was published. Clearly, the first author would have been told, and most likely also MTL.

"Senior leaders at Genentech including Dr. Tessier-Lavigne knew of the inconsistent binding results, and there was uncertainty and speculation within the Genentech Research organization about why the binding interaction between DR6 and N-APP could not be reliably reproduced or confirmed."

"Also following Dr. Tessier-Lavigne’s departure, one senior leader in gRED urged that the 2009 Nature paper should be retracted or corrected in light of the inconsistent binding results. Other senior leaders recognized at the time that this was an action only Dr. Tessier-Lavigne or another co-author could take with the journal." --> MTL was asked to retract and did not.

TLDR: the report is very damning. Why don't you try to dispute some of the facts reported by the Daily, instead of writing nebulously that their headline is misleading.

```

UPDATE: To clarify, that comment is responding to another comment saying "the report is very positive for MTL"

[1] https://stanforddaily.com/2023/04/06/stanford-president-rese...

[2] https://www.gene.com/download/pdf/Findings-of-2023-Genentech...

What does it count as to follow a bunch of racist tweeters? Are you saying it's not indicative of anything?

I didn't exactly claim that he said something objectionable. I did accuse him of broadcasting white supremacist tendencies by his follows. You're free to interpret the significance of these follows differently. I personally would be uncomfortable following accounts that retweet and tweet clearly racist statements. You can go to @empthyhaver for examples literally from the last couple of days.

I've never noticed him following any of the left-leaning accounts I know, and I've had tons of far right stuff that he follows recommended to me by Twitter after following him, so it seems like he's leaning one way.

Fair point, added.

> Can you go into more details about that? Who exactly are the "creepy ones" he's following on Twitter?

A while ago, someone wrote an article on it: https://theoutline.com/post/6708/marc-andreessen-twitter-fav...

But that's very out of date. I found out (before reading that) by following him, and then for a month my "for you" feed was full of cringy far right accounts (not Robert Spencer-level, but not too subtle either). Just some random ones I saw recently: @FistedFoucault, @empathyhaver, @0x49fa98