Viet Lai

Posts are my own.

[Viet Lai @laidacviet]

@aiwithadb No business doing business with the US gov dare to use DeepSeek API. They may use DeepSeek deployed on their own GPUs. The model is good enough for many things, but the API and trust in the owner of that API are not earned.

[Viet Lai @laidacviet]

@lqgist @AstroMikeMerri @arxiv arXiv wasn't meant to be guarded. People misunderstand the meaning of "pre-print".

[Viet Lai @laidacviet]

@i_mika_el @solopribuilds Giving birth?

[Viet Lai @laidacviet]

@kirillacharya I did that to one paper and properly on another paper I am reviewing. This mess should end.

[Viet Lai @laidacviet]

@ZhMlliang I think vibe coding is fine. I save a lot of time tbh. But producing a artifact for human that human cant understand is pointless. So as a reviewer, you need to guard to make sure people can understand the paper easily.

[Viet Lai @laidacviet]

I am so disappointed with the quality of the #ICLR papers. Some of them are just not worth spending time reading another paragraph, even if I tried to read them 3-4 times.

[Viet Lai @laidacviet]

Turns out if you write a proper paper, you immediately within top 20-40%.
AI slops make legitimate paper stand out

[Viet Lai @laidacviet]

@gsarti_ Good way to avoid bad paper, but if bad paper ends up in a not-good-gatekeeping reviewer, that's bad for the community, isn't it?

[Viet Lai @laidacviet]

@FaeghehSardari {CVPR, ICCV, and ECCV} {ICLR, ICML, NeurIPS} should team up and use the same review system; a paper just can't resubmit without substantial revision and declare that it is a resubmission. ACL ARR has done this and resubmissions face serious scrutiny from reviewers.

[Viet Lai @laidacviet]

@CSProfKGD 2/5 are OK-ish doses of AI. 3/5 are bad; I just want to desk-reject them immediately, and no reviewer needs to spend time on them.

[Viet Lai @laidacviet]

If you are interested in #internship at @AdobeResearch feel free to reach out and share your ideas and interests. (no promise atm).

[Viet Lai @laidacviet]

I am going to attend #COLM in San Francisco next week. Looking for people to talk to about vision-language models.

[Viet Lai @laidacviet]

@AdobeResearch I will present our new work in lost in the long conversation in AIW workshop at @COLM2026. The paper is accepted @ @NeurIPSConf , btw.

https://t.co/zCUnY0Nr0a

[Viet Lai @laidacviet]

@mehrdadhw @lihanc02 That cap is useless. I think 5 is serious and makes people really focus on the work that matters most to them and is of the highest quality. If other work is not ready or low quality, and significantly contributed

[arXiv.org @arxiv]

arXiv has updated our policy on rate limiting for all submitters.

This update was made to fairly distribute moderator time & support the arXiv community of staff, volunteers, readers & authors.

Please read our announcement to learn more: https://t.co/lvzWuiv0hx

[Trending AI Topics @aitrendings]

FlowTool: Controlling Tool Parameter in Image Retouching via Flow Matching

👥 Thanh-Long V. Le, Steven Walton, Seunghyun Yoon et al.

#AIResearch #ComputerVision #AI #MachineLearning

🔗 https://t.co/3sqPerKmsg

[Viet Lai @laidacviet]

@ahatamiz1 Its value comes from its smaller model weight. Some applications can afford compute but not the model's size.

[Viet Lai @laidacviet]

@Eng_Hemdi @klbouman I like her presentation in ICLR just a month back.

[Viet Lai @laidacviet]

@ChristosKyrkou @praeclarumjj Can I name? IBM

[Viet Lai @laidacviet]

@Maya64ce They said it. Literally as live as I am typigg ng now.

[Viet Lai @laidacviet]

CVPR oral presentation this year is a disaster. They read scripts, copy-pasted tables from paper and couldn't answer questions because they arent the people who did the work.

CVPR organizer should not let these talk at the precious time of thousands of researchers.