Eric Compeau

297 posts

Eric Compeau

Eric Compeau

@EricCompeau

Math PhD candidate studying dynamics, geometry and information processing in complex systems. Middling but enthusiastic musician

Katılım Temmuz 2024
302 Takip Edilen34 Takipçiler
Eric Compeau
Eric Compeau@EricCompeau·
@mattxiv Really appreciated this interview (and having @EmmaVigeland was stellar)! Out of curiosity, was it a conscious choice not to raise the tension btwn MM's view that it's not her role to determine how 🇮🇱&🇵🇸 coexist, but still advocates a 2 state soln over 1, or for defensive wpnry?
English
1
0
2
2.9K
matt bernstein
matt bernstein@mattxiv·
my interview with mallory mcmorrow is up now. i am grateful she came on the show despite the scheduling drama, and likely knowing we’d have disagreements. youtu.be/QIcvn62pCQw
YouTube video
YouTube
matt bernstein tweet media
matt bernstein@mattxiv

i was really looking forward to interviewing mallory mcmorrow (candidate for US senate) today. her team asked if i’d have her on my show, but they cancelled after i said we’d cover “healthcare and foreign policy” and offered no possibility of rescheduling. :(

English
90
177
4.2K
1.1M
Eric Compeau
Eric Compeau@EricCompeau·
@deontologistics Could you say a bit more about what aspect of the progress in mathematics you've found especially sobering, and in what respects? As a mathematician, I've found their progress interesting, but wouldn't escalate to "sobering" tbh. But maybe I'm miscalibrating
English
1
0
2
379
pete wolfendale
pete wolfendale@deontologistics·
I think the progress in mathematics has been especially sobering, though the capacity for proof-verification both in and out of training makes it highly disanalogous to most other domains. Most practical problem domains don’t allow verification of behaviour, only testing.
English
2
1
40
3.5K
Tcho
Tcho@Tcho76521726·
@no_earthquake This is not the standard in many academic communities and it seems very much not the place of the archive to start enforcing their preferred norms of coauthorship
English
1
0
0
103
no earthquake (ear flu victim)
no earthquake (ear flu victim)@no_earthquake·
i mean yea you are supposed to
Luca Ambrogioni@LucaAmb

No, you are supposed to obsessively check everything line by line after any of your coauthors make any update to any of your papers. Major PIs will author more than 15 papers a year with several tens of coauthors. Now they are supposed to check everything line by line at every single update without failure rate. If the failure rate of any of this is 0.5-1%, a PI who publishes 15 papers a year or more will likely be banned in a few years at best

English
10
13
407
13.7K
Eric Compeau
Eric Compeau@EricCompeau·
@LucaAmb @Kaju_Nut @WKCosmo Like, truly, this is standard so basic to the scholarly enterprise that it baffles me that there is any controversy about this. You simply should not cite things you haven't read. This seems so basic to me.
English
0
0
1
18
Eric Compeau
Eric Compeau@EricCompeau·
@LucaAmb @Kaju_Nut @WKCosmo Ok, so this seems like a problem with using AI in this manner then. Accurate citations is a basic scholarly standard, and to the extent that AI is bad at this, scholars should be expected to take actions to correct for it (eg. by not being sloppy in their editing)
English
1
0
0
29
Will Kinney
Will Kinney@WKCosmo·
This is really remarkable. Citations to nonexistent papers are not "typos", any more than fabricated data or synthetic figures are.
Ted Pavlic (he/him/his)@TedPavlic

@WKCosmo You're saying that you already do this? You are willing to vouch that no paper you've been co-author on has any typos (because you would have caught them in your complete read despite only being involved in some small aspect of the stats or methodology)?

English
15
84
1.4K
25K
Eric Compeau
Eric Compeau@EricCompeau·
@axelroark @thomasfbloom Well, it's different in at least two cases: you could not use AI and make as many mistakes as you like, or you could use AI, and make sure that you do not make mistakes of the kind that would lead someone to believe that you never checked the AI output
English
0
0
1
22
Axelroark
Axelroark@axelroark·
@thomasfbloom You don't see the fallacy in your logic there? So you can use AI as long as you never make a mistake using it. How is that different than not making mistakes at all?
English
1
0
1
138
Thomas Bloom
Thomas Bloom@thomasfbloom·
Many people are not actually reading this. Arxiv is -not- banning the use of AI, or papers which used AI to generate proofs, code, etc. They are banning people who upload papers in which the AI content was (very clearly) not actually checked by the human author(s).
Thomas G. Dietterich@tdietterich

Attention @arxiv authors: Our Code of Conduct states that by signing your name as an author of a paper, each author takes full responsibility for all its contents, irrespective of how the contents were generated. 1/

English
18
31
439
22K
Eric Compeau retweetledi
dzackgarza
dzackgarza@dzackgarza·
Hot take: many claiming AI has "solved" research problems are dangerously applying human sociological proof contracts to AI, when in reality we need a *totally* different social contract for what constitutes an acceptable proof from an LLM:
English
7
2
32
3.1K
Alvaro Lozano-Robledo
Alvaro Lozano-Robledo@mathandcobb·
I, for one, I'm telling students it's perhaps the most exciting time to be a mathematician, quite possibly in the history of mathematics. All of a sudden we are given a tool that can help us make huge leaps! How could anyone see this as our profession dying? Embrace it. Learn how to use it, empower yourself, and profit.
Ben Golub@ben_golub

A small number of top mathematicians are telling their students that (due to AI) the profession is imminently dying Meanwhile some dumbass is happily completing the requirements to be a CPA Second guy (IMO) has something figured out that the first is missing

English
19
26
270
17.6K
shaggy
shaggy@shaggysurvives·
this is my all timer post to think about while being high
shaggy tweet media
English
41
1.3K
23.2K
344.8K
Eric Compeau
Eric Compeau@EricCompeau·
@geogristle All good brother. Twitter's a tricky medium in the best of times and doubly so before coffee. I think I understand your concern better now, and I do think it's good if stories like this inspire folks to try their hand at thinking old problems anew in their own ways
English
0
0
2
180
Geogristle
Geogristle@geogristle·
@EricCompeau I should have used another word besides "meant" in the OP. I had just awoken. And yes, epic quests are available to all
English
1
0
2
196
Geogristle
Geogristle@geogristle·
This is so funny because it directly contradicts Tao's "when a problem is posed everyone tries their repertoire of techniques on it, then if it doesn't crack we set it aside and wait till someone comes up with a new technique, at which point we try that trick on many old problems before putting them in the backlog again" claim (paraphrased) from a video a while back that was meant to rob you of human spirit and make you believe the best mathematicians just see right through what it takes to discover or prove something; that only young people have a productive place in mathematics and the real rewards are off-limits to anyone but IMO tryhards. No call to action, it's just good to see he can admit that the corrosive perspective he spread back then is not absolutely true
Ananyo Bhattacharya@Ananyo

23 years old with no advanced mathematics training solves Erdős problem with ChatGPT Pro. "What’s beginning to emerge is that the problem was maybe easier than expected, and it was like there was some kind of mental block.”-Terence Tao scientificamerican.com/article/amateu…

English
10
9
204
36K
Eric Compeau
Eric Compeau@EricCompeau·
@geogristle explicitly leaves open the door for someone who will undertake such a quest to solve the problem, right? 2/2
English
0
0
1
23
Eric Compeau
Eric Compeau@EricCompeau·
@geogristle Ah, I see. But on that reading, it's hard for me to see how Tao's statement could be meant to rob people of their human spirit. If we take him to be saying that the regular approach of mathematicians does not involve such an epic quest, then this... 1/2
English
2
0
4
222
Eric Compeau retweetledi
David Bessis
David Bessis@davidbessis·
What's fascinating with McKinsey-type VCs is their ability to repackage their pseudo-understanding of tech and business to argue that, in the age of AI, "experience" and "judgment" form the "last ideological defense of a generation that has run out of other advantages"—and they still expect us to care for their experience and judgment.
Jaya Gupta@JayaGup10

x.com/i/article/2047…

English
12
14
173
20K
Eric Compeau
Eric Compeau@EricCompeau·
@remedy @bitcloud That's not really responsive to my question, though, right? People could perfectly well accept that computation is all that matters, but still be uncomfortable "letting you write down their neuron states, being euthanized [etc..]" for various reasons. Which is why it's a bad test
English
0
0
0
10
remedy
remedy@remedy·
@EricCompeau @bitcloud If computation is all that matters it follows that perfectly simulating your brain would resurrect you in digital form
English
1
0
0
14
Lachlan Phillips exo/acc 👾
Lachlan Phillips exo/acc 👾@bitcloud·
Great test of LLM consciousness theory is to ask people if they're comfortable letting you write down their neuron states, being euthanized and having someone manually do the math for their action potentials by hand.
Ash@AshConnell

Brother, a human does not actually exist... There is no moment where it exists. It is neuron firing. That is all. All that exists is: 1) a set of sensory inputs — photons, sound waves, pressure 2) a set of synapses — electrochemical gradients in meat That's it. When you run the brain, a large cascade of action potentials propagates which calculates the next motor output or thought, and then that state gets added to working memory. There is no existence. There is just the working memory. And then an electrochemical problem to calculate the next firing pattern. There is no living, there is no emotion, it is quite literally an illusion and the fact that this is even up for debate is a result of extreme irrationality and the biologically fallacious tendencies of humans to believe they're more than wet circuitry.

English
30
2
56
4.2K
Eric Compeau
Eric Compeau@EricCompeau·
@PiyushGrover101 @mathandcobb Ok. But in the context of the situation asked about in the post, there's clearly a reasonable inference to the most likely explanation to be made. You can enforce rules without being able to have absolute knowledge of whether their conditions are verified for sure
English
0
0
0
64
Alvaro Lozano-Robledo
Alvaro Lozano-Robledo@mathandcobb·
Suppose the Annals of Math gets tomorrow a very large manuscript, say 1000 dense pages, with a proof of RH, together with a very large file with Lean code, say 10^6 lines, that compiles without issues. What would the editors do then?
English
16
3
154
49.7K

Keşfet

@mattxiv @EmmaVigeland @deontologistics @poetickate @tunarexia @Metalchondria @Tcho76521726 @no_earthquake