Anas Awadalla (@anas_awadalla) 's Twitter Profile
Anas Awadalla

@anas_awadalla

phd student @uwcse | 🇵🇸

ID: 767263682300608512

linkhttps://anas-awadalla.streamlit.app calendar_today21-08-2016 07:34:51

214 Tweet

1,1K Followers

733 Following

Mehdi Hasan (@mehdirhasan) 's Twitter Profile Photo

Let’s be clear: Israel, an occupier, has no authority under international law to board or divert the Madleen. This is a hijacking, plain and simple. A hijacking of a UK-registered ship, with multiple *European* citizens on board. Anything to say Keir Starmer or Ursula von der Leyen?

Salesforce AI Research (@sfresearch) 's Twitter Profile Photo

🎉 BLIP3-KALE won Best Paper at the Synthetic Data for Computer Vision Workshop at #CVPR2025! 🎉 Congrats to the amazing team: Le Xue Le Xue, Manli Shu, An Yan, Silvio Savarese Silvio Savarese and Caiming Xiong Caiming Xiong! Really excited to see how this breakthrough is

🎉 BLIP3-KALE won Best Paper at the Synthetic Data for Computer Vision Workshop at #CVPR2025! 🎉

Congrats to the amazing team: Le Xue <a href="/Le_Xue01/">Le Xue</a>, Manli Shu, An Yan, Silvio Savarese <a href="/silviocinguetta/">Silvio Savarese</a> and Caiming Xiong <a href="/CaimingXiong/">Caiming Xiong</a>!

Really excited to see how this breakthrough is
Badr AlKhamissi (@bkhmsi) 's Twitter Profile Photo

🚨New Preprint!! Thrilled to share with you our latest work: “Mixture of Cognitive Reasoners”, a modular transformer architecture inspired by the brain’s functional networks: language, logic, social reasoning, and world knowledge. 1/ 🧵👇

🚨New Preprint!!

Thrilled to share with you our latest work: “Mixture of Cognitive Reasoners”, a modular transformer architecture inspired by the brain’s functional networks: language, logic, social reasoning, and world knowledge.

1/ 🧵👇
Thao Nguyen (@thao_nguyen26) 's Twitter Profile Photo

Web data, the “fossil fuel of AI”, is being exhausted. What’s next?🤔 We propose Recycling the Web to break the data wall of pretraining via grounded synthetic data. It is more effective than standard data filtering methods, even with multi-epoch repeats! arxiv.org/abs/2506.04689

Web data, the “fossil fuel of AI”, is being exhausted. What’s next?🤔
We propose Recycling the Web to break the data wall of pretraining via grounded synthetic data. It is more effective than standard data filtering methods, even with multi-epoch repeats!

arxiv.org/abs/2506.04689
Joongwon Kim (@danieljwkim) 's Twitter Profile Photo

Can we improve Llama 3’s reasoning abilities through post-training only? Introducing ASTRO, our new framework that teaches LLMs to perform in-context search and generate long CoT to solve math problems, via SFT and RL. Work done at @aiatmeta. 📄 Paper: arxiv.org/abs/2507.00417

Kevin Farhat (@notkevinfarhat) 's Twitter Profile Photo

The bottleneck in AI isn't just compute - it's access to diverse, high-quality data, much of which is locked away due to privacy, legal, or competitive concerns. What if there was a way to train better models collaboratively, without actually sharing your data? Introducing

The bottleneck in AI isn't just compute - it's access to diverse, high-quality data, much of which is locked away due to privacy, legal, or competitive concerns.

What if there was a way to train better models collaboratively, without actually sharing your data? 

Introducing
Yuhui Zhang (@zhang_yu_hui) 's Twitter Profile Photo

🧬 What if we could build a virtual cell to predict how it responds to drugs or genetic perturbations? Super excited to introduce CellFlux at #ICML2025 — an image generative model that simulates cellular morphological changes from microscopy images. yuhui-zh15.github.io/CellFlux/ 💡

Seth Bannon (@sethbannon) 's Twitter Profile Photo

There should be no place in tech for outright racism. Telling a Palestinian founder that his “friends” committed the October 7th massacre is exactly that.

There should be no place in tech for outright racism.

Telling a Palestinian founder that his “friends” committed the October 7th massacre is exactly that.
Alex Shaw (@alexgshaw) 's Twitter Profile Photo

Evaluating agents on benchmarks is a pain. Each benchmark comes with its own harness, scoring scripts, and environments and integrating can take days. We're introducing the Terminal-Bench dataset registry to solve this problem. Think of it as the npm of agent benchmarks. Now

Evaluating agents on benchmarks is a pain. Each benchmark comes with its own harness, scoring scripts, and environments and integrating can take days.

We're introducing the Terminal-Bench dataset registry to solve this problem. Think of it as the npm of agent benchmarks.

Now
Y Combinator (@ycombinator) 's Twitter Profile Photo

CopyCat is the AI RPA. Companies, from enterprises to startups, are saving thousands of hours building browser agents to automate gov sites, legacy portals, and dashboards, where APIs can’t. Congrats on the launch, Graham Sabin, zyad & Abhi Balijepalli! ycombinator.com/launches/O1l-c…

Zachary Foster (@_zachfoster) 's Twitter Profile Photo

The Gaza Humanitarian Foundation mercenary Anthony Aguilar testifies that the Israeli army is murdering innocent civilians at the so-called aid sites. Harrowing eye-witness testimony: “We treat these civilians in Gaza worse, with less dignity than we treated the ISIS fighters that surrendered at

Ro Khanna (@rokhanna) 's Twitter Profile Photo

I will not be intimidated. You don't know me or my grandfather. "Khanna...whose district has significant Israeli Americans.. could face significant backlash from the Bay Area Jewish community...Ro Khanna has disappointed many,"' warned Tali Kilma. jweekly.com/2025/08/01/rep…

Xiang Yue@ICLR2025🇸🇬 (@xiangyue96) 's Twitter Profile Photo

20 months after our multimodal reasoning benchmark MMMU (mmmu-benchmark.github.io) release, both frontier and open models are closing in fast but they’re still behind the ensemble of top human experts. (Made this figure last night after GPT-5 dropped—congrats on the new SoTA!)

20 months after our multimodal reasoning benchmark MMMU (mmmu-benchmark.github.io) release, both frontier and open models are closing in fast but they’re still behind the ensemble of top human experts. 

(Made this figure last night after GPT-5 dropped—congrats on the new SoTA!)
Omar Shaikh (@oshaikh13) 's Twitter Profile Photo

If you thought referencing past chats was cool, we built an MCP that lets Claude use *anything you see or do on your computer* as context.