@OmnAI_Lab

OmnAI lab, @SapienzaRoma, Computer Science Department (DI) PI: @_iAc

 Rome
omnai.di.uniroma1.it/
Joined October 2025
  • Tweets62
  • Following86
  • Followers70
2Media
We are at #ICLR2026 🇧🇷 presenting 5 papers spread across the main conference, 23-24-25 April. Stop by if you are interested in trustworthy and safe AI, generative models, robustness, and model inversion with @Hussain68018934 @BrigliaRosaria @adrianrminut Dario, and Hazem
6
17
391
OmnAI Lab retweeted
Paradigma
@paradigmaincOct 6
We are releasing the technical report for Limite 1B -Violetto. Violetto is our first model, designed to push mathematical capabilities per parameter to its limits, trained from scratch.
15
89
29
579
63,190
proud of my qt. owning the stage!
Ilir Aliu
@IlirAliu_Sep 28
Replying to @IlirAliu_
@paradigmainc Automated science. @tensorqt, @thelokasiffer, and Emanuele Rodolà are building AI systems that can participate in the research loop itself, from hypothesis generation through experimentation and validation.
1
2
32
1,840
OmnAI Lab retweeted
alex fazio
@alxfazioSep 24
emma, the italian meme slop model, made headlines in italy for months, but this didn’t. if i speak i'm in big trouble
Paradigma
@paradigmaincSep 21
Introducing Limite 1B - Violetto. A model for high-frequency mathematical intelligence.
24
32
1
498
23,240
OmnAI Lab retweeted
ItalAI
@_italaiSep 23
Our Co-founder & CTO Fabio Galasso @GalassoFab10 and General Manager Luca Franco @FraLuca17 are at VISMAC today and tomorrow. If you're attending, feel free to reach out! #VISMAC26
1
9
240
our first model redefines what is possible under strict parameter budgets, and addressing mathematical reasoning felt timely and natural we'll be serving many more colors soon!
Paradigma
@paradigmaincSep 21
Introducing Limite 1B - Violetto. A model for high-frequency mathematical intelligence.
4
4
103
4,815
OmnAI Lab retweeted
Paradigma
@paradigmaincSep 21
Introducing Limite 1B - Violetto. A model for high-frequency mathematical intelligence.
64
226
78
1,830
289,189
OmnAI Lab retweeted
Paradigma
@paradigmaincSep 18
8
10
3
72
7,808
Nihar Shah did a heroic experiment for TMLR: he spent 20-25 hours over two weeks interviewing authors of seemingly low-quality submissions about their own papers. He confirmed what we all suspected: people submitting these papers have *no idea* what is going on in them.
Transactions on Machine Learning Research@TmlrOrgSep 16
TMLR has faced a deluge of submissions, necessitating stricter desk rejection policies due to limited reviewer capacity Co-EiC Nihar Shah reached out to authors of 10 papers slated for desk reject. Could they answer questions about their *own* submission? medium.com/@TmlrOrg/asking-a…

Asking Authors About Their Own Papers

By Nihar B. Shah

medium.com
12
138
27
1,312
175,298
OmnAI Lab retweeted
Paradigma
@paradigmaincAug 12
Automating research means pushing boundaries. A retrospective on the process of doing it. Authored by our Panisperna Fellows.

A Retrospective on Our World Records

Article · @paradigmainc
2
5
5
38
9,299
OmnAI Lab retweeted
Paradigma
@paradigmaincAug 10
Some recent work by our Panisperna Fellows, more is coming soon.
Larry Dial
@classiclarrydAug 9
New NanoGPT Speedrun WR at 73.8s (-0.8s) from Mister-dev-oss, CerovazS, MarioPaerle, GabrieleCirillo, and crisostomi on GitHub. They added an incredibly sophisticated fp8 implementation on the MLP down proj fwd pass. 280 lines that include putting the activation quantization step into the prior kernel, delayed amax scaling, and fusing the weight quantization and transpose together. github.com/KellerJordan/modd…

Track 1: Add FP8 MLP down projection (-1.04s) by Mister-dev-oss · Pull Request #342 · KellerJorda...

Summary This PR builds on the current Track 1 master record and adds one systems-side change to train_gpt.py and triton_kernels.py: FP8 MLP down projection. The MLP down-projection forward path now...

github.com
5
21
4,003
Vision conferences do this well. Fixes for NeurIPS. - Remove the discussion period. It's useless. - 1 page rebuttal, no LLM discussions. Everyone can read. - Separate preliminary and final rating. Desk reject papers of reviewers who do not fill in final rating. Let's start here.
9
12
1
182
19,668
OmnAI Lab retweeted
Paradigma
@paradigmaincJul 17
We are proud to announce we have been selected as a challenger in the Next Frontier AI programme. Europe is behind. It took us a while to realize it, and a little longer to do something about it. But there’s one key thing we’ve been behind on, and it’s neither compute nor capital, but rather the culture that’s required to attempt impossible feats and come out victorious. This challenge, finally, proves otherwise.   The road in front of us is extremely difficult and uncertain, the promise is to save our continent from irrelevance and to usher in a future of prosperity.  We were born for this. If a challenge this difficult excites you instead of depressing you, if it fills you with the urge to open up new frontiers instead of despair, if you think Europe isn’t done, we want to talk to you.
16
10
10
94
13,719
OmnAI Lab retweeted
ItalAI
@_italaiJul 13
20 years of ICVSS 🦾 Computer Vision for Spatial & Physical Intelligence
2
6
243
📢 A quick reminder that the Unlearning and Model Editing (U&ME) Workshop @ ECCV 2026 is still accepting submissions. ⏳ Deadline: 9 July (AoE) 📌 Details: sites.google.com/view/u-and-… @eccvconf #ECCV
1
1
2
755
OmnAI Lab retweeted
Xudong Han
@Xudong07452910Jun 9
这可能是人类写给 AI 看的最后一篇论文了。 最近刷到Stanford、CMU、Michigan 等 37 位作者联名的论文：《The Last Human-Written Paper》。 核心观点很狠：沿用几百年的论文，在 AI 时代可能已经过时了。 作者点出了两个被我们忽视已久的“隐形税”： 一个是叙事税。为了讲一个漂亮故事，我们把失败实验、死路、被推翻的假设都删掉了。AI 读到的是“通关攻略”，却看不到真正有价值的“踩坑记录”。 另一个是工程税。论文里的实现细节通常足够说服审稿人，但不够让 Agent 直接复现。很多关键 tricks 还藏在作者脑子、代码注释和 Slack 记录里。 所以作者提出 ARA，直接把论文改造成 Agent 能读取和执行的“研究包”：不只告诉你结论，还把怎么想到的、代码怎么跑、证据链在哪、哪些路走不通都打包进去。 我觉得这篇最有意思的地方是，它不是在讨论 AI 怎么帮人写论文，而是在问： 当 AI 也变成论文读者和执行者时，论文还应该长成今天这样吗？ 未来科研输出的核心，可能不再是“写得多像一篇 paper”，而是能不能被 AI 理解、复现、追踪和继续扩展。 人类写论文写了几百年，接下来可能要开始写给 Agent 执行的研究包了。 arxiv.org/pdf/2604.24658
134
494
69
2,225
248,520
OmnAI Lab retweeted
ItalAI
@_italaiJun 5
Thanks to everyone who joined us at the MUV Workshop at @CVPR ! @muv_workshop focused on selective data removal, concept erasure, debiasing & model editing, as well as safe image/video generation & post-training control. Great to see the community come together to advance the future of safe, privacy-aware, and trustworthy AI. #CVPR2026
1
11
264
📢CPF: U&ME Workshop @ ECCV 2026 @eccvconf We invite submissions on machine unlearning, model editing, and related topics including efficient adaptation, and responsible AI. Details at: sites.google.com/view/u-and-… @_iAc #ECCV #ECCV2026 #Unlearning #AIsafety #ResponsibleAI
🤖 Made with AI
3
2
5
3,089
Load more