3.5 yrs ago, I faced the decision of pursuing a startup or a PhD. I chose later Turns out, PhD is a series of multiple 6-8 months startup. You find an idea, prove it works, find collaborators who compliments you, producing results, present it in a simple way. Start again

The results are really cool! But one unpopular opinion: Simulating a browser inside a separate web app deviates from how users naturally browse. Web agents should integrate seamlessly—like a Chrome extension—not in isolated sandboxes. That’s the approach we took in CowPilot.

I interviewed for LLM/ML research scientist/engineering positions last Fall. Over 200 applications, 100 interviews, many rejections & some offers later, I decided to write the process down, along with the resources I used. Links to the process & resources in the following tweets

🚀 Rising Star Workshops for Junior/Senior PhDs, and Postdocs! 🌟 Don't miss these career-boosting opportunities! notion.so/List-of-Rising… Please share with your peers, students, and anyone who might benefit! #PhD #Postdoc #Academia #RisingStars

Excited to share that our CowPilot🐮 is accepted to #NAACL 2025 Demo Track! Definitely check out our user study if you're interested in trying out CowPilot: forms.gle/aYkXhh7fdRwZ94…

what if an agent's activity was more visual and understandable

📢 Join us at the CMU Agent Workshop 2025, April 10-11! Don't miss our esteemed invited speakers: - Qingyun Wu (PSU) - Diyi Yang (Stanford) - Aviral Kumar (CMU) - Graham Neubig (CMU) ...and many more to come! To register, visit: cmu-agent-workshop.github.io

people don't talk enough about how pretty Anthropic's brand really is

Only five days remain until the poster submission deadline for the CMU Agent Workshop 2025! 📷 Submission link: cmu-agent-workshop.github.io/posters

A new AI agent from the startup Simular switches between different AI models depending on the task at hand. wired.com/story/simular-…

Thanks Will Knight for featuring CowPilot in WIRED! He explains how agents can benefit from human intervention in a collaborative setting, leading to better performance. Stay tuned for more updates on CowPilot in the coming days! wired.com/story/simular-…

Excited to share my first project at UCLA! We built MOSAIC — a social network simulator where LLM-powered agents behave like real users on social media. They post, share, flag, and debate the factuality of news content — all at scale. It’s open-source. 🧵 TL;DR 🌐 Realistic

Just found out about a recent startup called Granola: granola.ai Seeing how NoTeeline and Granola have the shared vision of maximizing the short/micronotes people take is impressive. Instead of letting AI take the whole note, just use it to make your notes better!

I will be at #NAACL2025 with my cat 🐈, presenting three works: - Apr 30: CodeRAG-Bench co-led with Akari Asai: code-rag-bench.github.io - May 2: CowPilot led by Faria Huq | 🦋: fariahuqoaishi: oaishi.github.io/cowpilot.html - May 2: Fail-TaLMs led by Eduardo Treviño and Hugo Contant, that benchmark

Amanda Bertsch @NAACL Simran Khanuja Lindia Tjuatja We also have a demo on CowPilot, a human-agent collaborative framework for web navigation. arxiv.org/abs/2501.16609

🚨 New Paper! 🚨 "Groups that communicate diverse information make wiser decisions." Right? Not always—for LLMs, just like humans. We bring this scrutiny to AI by adapting the Hidden Profile paradigm from social psychology to assess how multi-agent LLMs actually reason

🚨 New Preprint! 🚨 Smarter LLMs are more selfish. We show reasoning-enhanced models significantly prefer greed over cooperation. The more LLMs reason, the worse they cooperate. 👇 THREAD 👇 [Link to paper: arxiv.org/abs/2502.17720] [1/n]

Struggling to process this --- I’m heartbroken. I was supposed to meet a brilliant Bangladeshi MIT PhD candidate yesterday morning. Chatted just a few days ago. Less than an hour before our meeting, I learned he is no longer with us in this world. He had strong publications, a