Arnaud Deza

Applied Scientist @Amazon SCOT | PhD student @georgiatech @mlatgt @AI4OPT Prev: @Salesforce @Huawei EngSci 2T2 @UofT

[Kosta Derpanis @CSProfKGD]

When looking for a PhD supervisor, prioritize strong supervision rather than simply chasing prestige.

Supervision done wrong 👇

[Arnaud Deza @CombinatorialAI]

The students publishing JEV papers 5 days after release should obviously be embarrassed, but the advisors are somehow worse. Wtf are you supervising if your student starts a paper days before the deadline and you just slap your name on it? Did you even read the thing?

[(((ل()(ل() 'yoav))))👾 @yoavgo]

JEV was released in "limited early access" on September 15, that's two weeks ago. How come we have 29 papers about it on arxiv (and, I assume, ICLR submissions) already. This is not healthy.

[Arnaud Deza @CombinatorialAI]

Post-ICLR cluster availability is a beautiful thing

[Arnaud Deza @CombinatorialAI]

I feel slightly less like a chimp after seeing this. Voice-to-text has me feeling like a smoothbrain in the office

[Morgan Grenfell @Corruptario4eva]

He has so much to offer to the job. 😂

[Arnaud Deza @CombinatorialAI]

@mmaaz_98 https://t.co/CLV9l4Fx9H lots of engineering work + figuring out how to run this on an academic cluster where the limit is 32 gpus/user; I go up to 76 :)

[Arnaud Deza @CombinatorialAI]

Arxiv submitted. I can finally go to sleep…

[Arnaud Deza @CombinatorialAI]

What is the largest linear program solved? Is it still google research with PDLP?

[Arnaud Deza @CombinatorialAI]

@suchenzang wow, the plot thickens...

[Pierre Chambon @PierreChambon6]

Can I tweet something about my updated arxiv or now is not a good time maybe ?

[Arnaud Deza @CombinatorialAI]

@RadishHarmers Banger

[Peyman Milanfar @docmilanfar]

when you’re getting a hard time at your poster and your advisor walks up

[beginbot 🃏 @beginbot]

sub agents being released into my codebase

[Arnaud Deza @CombinatorialAI]

@alec_helbling Wow! Have you tested any other discrete optimization problem than sudoku? I.e doing this for maxcut or vertex cover etc I.e neural combinatorial optimization

[Arnaud Deza @CombinatorialAI]

@mmaaz_98 Hall of fame tweet! Still remember when lava's was 2.25 CAD for a slice bigger than your whole face

[Arnaud Deza @CombinatorialAI]

@marshallfrith @theiaincameron @grok 🐐🙏

[Arnaud Deza @CombinatorialAI]

@marshallfrith @theiaincameron super great work! How do you do this? Is there software for this that is free or do you have to pay for some sort of API to get the satellites snapshots?

[Arnaud Deza @CombinatorialAI]

What is the largest linear program solved? Is it still google research with PDLP?

[Arnaud Deza @CombinatorialAI]

I feel like it's abit nuanced as, to do this day, it seems most RS/AS/MTS in top labs and FAANG do have PhD's and were able to intern at top labs because they were PhD students...

[Nathan Lambert @natolambert]

I tell a ton of people that getting a PhD is not a good idea for your career, but it’ll likely be far higher in personal growth and fulfillment which is maybe worth way more anyways. I need to think about how to price that in.

[Maaz @mmaaz_98]

I lowkey had a crisis of faith in academia in 2024 when I realized the vast majority of “novel” ML methods, esp from famous profs, were using poorly tuned baselines. I remember I spent hours on the phone complaining about it to my friends!

[Patrick OShaughnessy @patrick_oshag]

Someone I trust, and who would know, thinks there’s max two years left of kernels engineering (which is, in many places, a major bottleneck today)

Late in his life Munger said “If only Warren and I could watch AI unfold. Not even participate. Just watch”

We all get to watch