←
Anthony Peng245 posts
Anthony Peng avatar

Anthony Peng

@RealAnthonyPeng

CS PhD @GeorgiaTech @PoloDataClub | Intern @nvidia, @Meta, @IBMResearch, @intel | Outcomes are what count; don’t let good processes excuse bad results.

◷ Joined January 2021▧ 45 Photos and videos
Anthony Peng avatar
Anthony Peng@RealAnthonyPeng·Aug 22

🎉 Excited to share that RECAP has been accepted to #EMNLP2026 Main! The key idea: Flawed thinking can actually help reasoning models learn better alignment. 🧠🛡️ We find that injecting just a small amount of flawed reasoning can reduce model safety by up to 36%. 📉 Instead of simply avoiding these failures, we train reasoning models to recognize, override, and recover from them. 💪 Introducing RECAP 🔄 — an RL post-training method that teaches reasoning models to overcome unsafe reasoning trajectories and reroute toward safe & helpful answers, without additional training cost. ✨ Safer reasoning 🔒 Stronger jailbreak resistance 🙅 Lower overrefusal 🧠 Preserved reasoning ability A huge thank you to all of our amazing collaborators who made this work possible! 🙌 I’m especially grateful to @PoloChau , @jianfengchi , and @pinyuchenTW for their invaluable support, feedback, and guidance throughout this project. #LLM #AISafety #AIAlignment

Post attachmentPost attachment
Anthony Peng avatar
Anthony Peng@RealAnthonyPeng·12h

I successfully defended my PhD at Georgia Tech! 🎓🐝🎉 𝘽𝙚𝙮𝙤𝙣𝙙 𝘾𝙖𝙥𝙖𝙗𝙞𝙡𝙞𝙩𝙮: 𝙎𝙖𝙛𝙚 & 𝙍𝙚𝙡𝙞𝙖𝙗𝙡𝙚 𝙁𝙤𝙪𝙣𝙙𝙖𝙩𝙞𝙤𝙣 𝙈𝙤𝙙𝙚𝙡𝙨 𝙖𝙩 𝙎𝙘𝙖𝙡𝙚 “Beyond Capability” was the title. “Beyond relieved” is the current status. 😮‍💨🥂 My research brings together AI safety 🛡️, robust model design 🏗️, and better use of data and compute 📈. Huge thanks to my advisor @PoloChau and committee @cocoweixu @alan_ritter @pinyuchenTW @jianfengchi for your mentorship and support! 🙏❤️

Post attachment
Anthony Peng avatar
Anthony Peng@RealAnthonyPeng·Nov 13

As an Nvidian, Jensen stepping down still feels surreal. I’m deeply saddened, and still trying to process it. It’s been hard to watch how easily speculation becomes certainty, and real people’s lives become entertainment. Behind the headlines are people and families living through this. I wish there were more room for empathy. I keep thinking about Jensen’s words on resilience, and how hardship shapes who we become. Those words give me something to hold on to right now. Wishing strength to everyone affected, especially my fellow Nvidians. Take care of each other. #Nvidia

Ben Hoover avatar
⟳ Anthony Peng retweeted
Ben Hoover@Ben_Hoov·Nov 13

Big reasoning speedups are here. 1,000× faster training and 280× less compute than any looped model. All done using a fully trainable energy from the 80s Keep reading to hear how energy is the right paradigm for reasoning With @HughBlayney @jcrwhittington @prefrontal_ai 🧵/10

0:25
Alex Yang avatar
⟳ Anthony Peng retweeted
Alex Yang@AlexanderHYang·Nov 9

Here's this year’s lineup of papers, workshops, doctoral colloquium presentations, and more from @GeorgiaTech at @ieeevis #IEEEVIS2026! 🎉 Check out the schedule below—including our Best Short Paper, Honorable Mention winner—and come say hello!

Post attachment
GaTech CSE avatar
⟳ Anthony Peng retweeted
GaTech CSE@GTCSE·Oct 26

3️⃣ papers. 2️⃣ CSE Ph.D. students. 1️⃣ major #AI research conference! 🤖 Anthony Peng and Ahren Jin are presenting this week at #EMNLP2026, exploring safer AI reasoning, multi-perspective AI companions, and better cultural understanding! Congrats, Anthony and Ahren! 🐝🥳 🔗🧵👇

Post attachment
Anthony Peng avatar
Anthony Peng@RealAnthonyPeng·Oct 25

I'll be presenting our paper, "Large Reasoning Models Learn Better Alignment from Flawed Thinking," at #EMNLP2026 today! ⏰ 10 AM PT / 1 PM ET / 6 PM CET 📍 ML.2a — Machine Learning for NLP Come join my talk! Looking forward to the discussions!

Frontier AI labs are talking more than ever about 𝐀𝐈 𝐬𝐚𝐟𝐞𝐭𝐲 🛡️ while releasing models faster than ever 🚀 We don’t release frontier models. But we do take their safety seriously 😄 Our 𝐄𝐌𝐍𝐋𝐏 𝟐𝟎𝟐𝟔 Main paper, 𝐑𝐄𝐂𝐀𝐏, studies how flawed reasoning can undermine model safety, and how to train reasoning models to recognize, override, and recover from unsafe trajectories 🔄🤖 Made a short video explaining the idea 🎬 Huge thanks to @pinyuchenTW, Eric Smith, @songjiang24, Hongyuan Zhan, @haozhu_wang, @mapa4u, @PoloChau, and @jianfengchi 🙌 #AISafety #LLM #EMNLP2026

10:46
GaTech CSE avatar
⟳ Anthony Peng retweeted
GaTech CSE@GTCSE·Oct 20

Congrats to CSE Prof. @PoloChau and his lab on being selected for the NVIDIA Academic Grant Program! 🥳 With new workstation GPUs in hand, the group is better equipped to advance research in robust AI! 🛠️ Many thanks to our generous supporters from @nvidia through #NVIDIAGrant!

Post attachment
Anthony Peng avatar
Anthony Peng@RealAnthonyPeng·Oct 14

Watching @karpathy teach LLMs is like watching MJ play basketball: he makes something ridiculously hard look ridiculously easy. Every video has me thinking, “Wait… I could build that.” That kind of clarity is a superpower. 🐐

Anthony Peng avatar
Anthony Peng@RealAnthonyPeng·Oct 7

Frontier AI labs are talking more than ever about 𝐀𝐈 𝐬𝐚𝐟𝐞𝐭𝐲 🛡️ while releasing models faster than ever 🚀 We don’t release frontier models. But we do take their safety seriously 😄 Our 𝐄𝐌𝐍𝐋𝐏 𝟐𝟎𝟐𝟔 Main paper, 𝐑𝐄𝐂𝐀𝐏, studies how flawed reasoning can undermine model safety, and how to train reasoning models to recognize, override, and recover from unsafe trajectories 🔄🤖 Made a short video explaining the idea 🎬 Huge thanks to @pinyuchenTW, Eric Smith, @songjiang24, Hongyuan Zhan, @haozhu_wang, @mapa4u, @PoloChau, and @jianfengchi 🙌 #AISafety #LLM #EMNLP2026

10:46
Alex Yang avatar
⟳ Anthony Peng retweeted
Alex Yang@AlexanderHYang·Oct 7

Excited to share our work at #ISMAR2026 🇮🇹! Our paper, “CrossAtlas: Evaluating Projection Techniques for Spatial Referencing in Cross-Reality Collaboration,” has been accepted to the IEEE TVCG journal track! ✨

1:00
Anthony Peng avatar
Anthony Peng@RealAnthonyPeng·Sep 30

Apparently spatial intelligence is worth 𝟖.𝟐𝐁 now. We settled for a 𝐍𝐞𝐮𝐫𝐈𝐏𝐒 acceptance 😄 Excited that 𝐕𝐄𝐑𝐈𝐖𝐎𝐑𝐋𝐃 was accepted to the 𝐍𝐞𝐮𝐫𝐈𝐏𝐒 𝟐𝟎𝟐𝟔 E&D Track 🎉 We study how VLMs perceive, reason, and act in interactive 3D worlds 🌐. We find that they often reason well once spatial structure is given explicitly, but still struggle to extract that structure from visual input 👀. We also find that the interaction setup itself can significantly change the outcome ⚙️. Big thanks to @YanZhengtexas & Atlas Wang 🙌 More soon! #NeurIPS2026 #SpatialIntelligence #VLMs

Post attachment
Peyman Milanfar avatar
⟳ Anthony Peng retweeted
Peyman Milanfar@docmilanfar·Sep 24

ICLR submissions

0:16