Elfarouk Harb 🌱

133 posts

Elfarouk Harb 🌱

Elfarouk Harb 🌱

@ElfaroukHarb

CS PhD Student at UIUC. I focus on algorithm design. Incoming Quant Researcher @twosigma. Previously SWE @Google and Quant Trader @citsecurities

Champaign, IL Katılım Eylül 2017
54 Takip Edilen62 Takipçiler
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
@aireteryx @peterepetepeter @fkasummer Also, Jensen came off incredibly defensive and kept regurgitating PR talking points. I get it from a stakeholder-responsibility angle, but it wasn’t a great look. The question was fair, and the uranium analogy, while imperfect, was at least gesturing at a real second-order effect
English
0
0
1
99
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
@aireteryx @peterepetepeter @fkasummer AFAIK Chinese labs’ creativity has been squeezing more of Nvidia chips, not replacing it. CUDA took 2 decades, the path isn’t purely horsepower, but horsepower is what trains the AI that shortens every other step. Chicken and egg, and Jensen is proposing they hand the egg.
English
1
0
1
119
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
@aireteryx @peterepetepeter @fkasummer Every H100 shipped to China trains models that accelerate their AI research, including research aimed at building the chip that replaces it. NVDA isn’t creating dependence, it’s financing their exit. The delay is the whole point, and why the uranium bit wasn’t complete dog water
English
2
0
3
327
Aire
Aire@aireteryx·
@peterepetepeter @fkasummer China won’t hustle to outcompete if they can just buy Nvidia chips but if not they have the will and capacity to do so. Any foreign dependence on a domestic product is good for us in aggregate. The uranium analogy is dog water
English
1
0
8
1.1K
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
@JFPuget The “vision” was that the CCP will be asked nicely not to militarize american frontier chips, got it. Dwarkesh can’t compete, reading blogs sets a low ceiling. Nobody said your employer dictates your thoughts, it’s just a hell of a coincidence which half you found profound 😅
English
0
0
1
224
JFPuget 🇺🇦🇨🇦🇬🇱
JFPuget 🇺🇦🇨🇦🇬🇱@JFPuget·
What is interesting is that you can't even parse one sentence till its end. One states his vision, the other one repeats what was written by someone else. Is that clearer? And if you think my employer dictates what I think then I feel for you. The planet you think you live in must be quite unpleasant.
English
1
0
1
1.7K
JFPuget 🇺🇦🇨🇦🇬🇱
JFPuget 🇺🇦🇨🇦🇬🇱@JFPuget·
My take on Dwarkesh - Jensen interview. One is a visionary, the other one is regurgitating the latest blog he liked.
English
54
20
794
112.9K
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
@JFPuget Genuinely cannot understand how a sane person watches Jensen’s china answer and decides *Dwarkesh* was the one regurgitating talking points…
English
2
0
6
2K
Elfarouk Harb 🌱 retweetledi
Chubby♨️
Chubby♨️@kimmonismus·
Yeah Claude. I know that feel. Wow.
Chubby♨️ tweet media
English
63
931
25.8K
449.8K
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
@AmeliaBarty Fucking hilarious that you think the cure for cancer won’t come with a terms of service agreement and a credit check. Who’s paying for it, the people without “utility”?
English
0
0
1
17
Amelia Bartholomew
Amelia Bartholomew@AmeliaBarty·
Honestly, what an incredibly tone-deaf thing to say Only someone with no experience of chronic disease, coupled with the hubris of the Silicon Valley tech elite, could have the audacity to hit send on this drivel This guy is a “Member of Technical Staff” at OpenAI btw. Had to do a double take because I couldn’t believe what I was reading
Amelia Bartholomew tweet media
English
14
6
118
18.5K
Maaz
Maaz@mmaaz_98·
@ElfaroukHarb This either cuts down on the dimension, or introduces additional affine constraints by possibly tying coeffs together, eg the template ax^2 + bx^2. For what it’s worth, it didn’t have any issues with Motzkin and other polys, but maybe I didn’t push it hard enough.
English
1
0
1
54
Maaz
Maaz@mmaaz_98·
I am finally releasing my new Python+Lean package, sostactic, for proving polynomial inequalities using sum-of-squares (SOS) certificates. It is significantly more powerful than current Lean tactics like nlinarith and positivity, and in practice much faster than an SMT solver.
Maaz tweet mediaMaaz tweet media
English
8
38
328
21.1K
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
@mmaaz_98 ... the solver could run for 10k+ iterations without really converging. So I’m wondering how do you choose the iteration cap in your pipeline, and how do you decide that the numerical solution is "good enough" to pass to the exactification step?
English
1
0
1
30
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
@mmaaz_98 Very nice writeup. One thing I’m curious about is the numerical side of the SDP solve itself. When I wrote an SOS checker using MOSEK through CVXPY and SDPs, numerical instability was an issue for me on some examples, especially Motzkin-type polynomials / ratio-of-SOS situations
English
1
0
2
163
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
@sheriyuo Academia loves identifying problems in the abstract, but the minute it has to attach them to actual people, everyone hides behind process. Put names on reviews, name and shame bad-faith actors, maybe even yearly shitty reviewer awards... but alas, that would require a backbone.
English
0
0
3
385
Xiuyu Li
Xiuyu Li@sheriyuo·
Got hit with a 2 score reviewer at ICML who clearly did not read the paper carefully. We wrote a 5000 word rebuttal and got back a single line saying thanks for the reply, however. There really should be some kind of reviewer reputation or credit system. Too many people seem to treat reviewing like a place to take out their frustration after their own papers get rejected.
English
8
5
238
37K
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
I never get tired of out-of-touch academicians sneering at industries that are at least trying to build tools for scientists to actually use, rather than constructing gated gardens of inaccessibility and calling it rigor. Self-reflection is truly a research frontier
English
0
0
0
48
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
Here at topconference.org, we accelerate science by ensuring only work blessed by the proper venue brands is allowed to count. Unlike the vulgar world, where truth is judged by being right, academia knows correctness requires 24 months, page limits, and one rejection first
Daniel Litt@littmath

Here at [famous scientist’s name].ai, we’re developing tools to accelerate science. Unlike academia, which has stifled the production of high quality scientific work by demanding it be correct, here at [fsn].ai, we know that you can just do (wrong) things.

English
1
0
2
116
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
@notadampaul @allgarbled @ludwigABAP This is what happens when you confuse verification with proof. Nobody doing serious math thinks a proof is just a certificate that a statement is true. The proof is where all the ideas live
English
1
0
3
57
notadampaul
notadampaul@notadampaul·
I do think that proof for its own sake is not nearly as valuable as people make it out to be, and that "knowing a thing is true" is rarely what actually advances mathematics. Or at least, while necessary, it is not sufficient I am not so bearish on autoformalization as ludwig but I do think some cultural pushback is beginning to come due
English
1
0
11
538
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
"can the AI do math alone?", it's what effective users are doing to make the collaboration work. A gold standard benchmark should study that, not rule it out. Requiring full transcripts would address any integrity concerns and preserve workflows that actually produce results.
English
0
0
0
37
Elfarouk Harb 🌱
Elfarouk Harb 🌱@ElfaroukHarb·
I think prohibiting expert-model interaction is misguided. Nearly every meaningful result so far has come from close human-AI collaboration, not autonomy. With current models, open or closed, pure autonomy feels set up to fall short. The most interesting question isn't ...
Tony Feng@tonylfeng

The second round of #FirstProof is announced: 1stproof.org. Glad to see the organizers planning a rigorous process. Most AI benchmarks today still rely on the "honor system". That’s billions of valuation dollars hinging on integrity, even while there are many performance-boosting tricks that are undetectable in results (adding hints to prompts, human intervention, best-of-N, etc). For this reason, one should interpret one-off performances by proprietary models (e.g., on math contests) with a grain of salt–especially when they are carried out after solutions are already posted online. FirstProof will be the only evaluation based on the natural distribution of research mathematics that guarantees autonomy and transparency of results. It should be considered the gold standard benchmark in the AI4Math space.

English
1
0
0
72
Elfarouk Harb 🌱 retweetledi
Zvi Mowshowitz
Zvi Mowshowitz@TheZvi·
My son is asked to find surface area of a cube for his homework. There is exactly one wrong side length for such a problem, and they picked it.
English
123
51
18.1K
1.7M
Elfarouk Harb 🌱 retweetledi
Danny Polishchuk
Danny Polishchuk@Dannyjokes·
Hey did you accidentally bomb Iraq instead of Iran??? ChatGPT: You’re absolutely right good catch!
Sam Altman@sama

Tonight, we reached an agreement with the Department of War to deploy our models in their classified network. In all of our interactions, the DoW displayed a deep respect for safety and a desire to partner to achieve the best possible outcome. AI safety and wide distribution of benefits are the core of our mission. Two of our most important safety principles are prohibitions on domestic mass surveillance and human responsibility for the use of force, including for autonomous weapon systems. The DoW agrees with these principles, reflects them in law and policy, and we put them into our agreement. We also will build technical safeguards to ensure our models behave as they should, which the DoW also wanted. We will deploy FDEs to help with our models and to ensure their safety, we will deploy on cloud networks only. We are asking the DoW to offer these same terms to all AI companies, which in our opinion we think everyone should be willing to accept. We have expressed our strong desire to see things de-escalate away from legal and governmental actions and towards reasonable agreements. We remain committed to serve all of humanity as best we can. The world is a complicated, messy, and sometimes dangerous place.

English
292
4.6K
65.6K
3.2M

Keşfet

@aireteryx @peterepetepeter @fkasummer @JFPuget @AmeliaBarty @mmaaz_98 @sheriyuo @elonmusk