「AIも使い続けていると人間のように老化する」という指摘、セッションを重ねて記憶が蓄積されることで性能悪化 https://gigazine.net/news/20260528-agingbench-ai-agents-age-too/

「AIも使い続けていると人間のように老化する」という指摘、セッションを重ねて記憶が蓄積されることで性能悪化

AIエージェントはどれだけ使用しても人間のように疲れることがなく、加齢などによる認知機能の低下もないと考えられています。ところが、テキサス大学オースティン校の研究チームが「AIエージェントも老化する」と指摘し、AIエージェントの老化を測定するベンチマークの「AgingBench」を発表しました。

gigazine.net

Congratulations to UT Computer Science Ph.D. student Yeonju Ro on being named a 2026 MLCommons ML and Systems Rising Star! Co-advised by Aditya Akella and @atlaswang.bsky.social, Yeonju is also a contributor to UT’s @ldosexpedition.bsky.social and Infra AI Center.

A new study shows that feeding large language models low-quality, high-engagement content from social media lowers their cognitive abilities. www.wired.com/story/ai-mod...

AI Models Get Brain Rot, Too

A new study shows that feeding large language models low-quality, high-engagement content from social media lowers their cognitive abilities.

wired.com

LLMs can get “brain rot”! www.linkedin.com/posts/atlas-...

Pleasantly surprised to see a side project, co-led by the now-professor Junyuan Hong and sparked by a totally random idea myself threw out back in January, gaining lots of (organic) traction on… | Atl...

Pleasantly surprised to see a side project, co-led by the now-professor Junyuan Hong and sparked by a totally random idea myself threw out back in January, gaining lots of (organic) traction on Twitte...

linkedin.com

Calling all parsimony and learning researchers 🚨🚨 The 3rd annual CPAL will be held in Tübingen Germany March 23–26, 2026! Check out this year's website for all the details cpal.cc

Got a comment I secretly like: “This professor is constructively rude”

www.linkedin.com/posts/atlas-...

Job Description | XTX Markets | Atlas Wang

🌐 It has been an incredible first year for the XTY AI Lab of XTX Markets in New York City, where we set out to build a world-class hub for AI research in quantitative finance: the results have exceede...

linkedin.com

Paying for a dinner I wasn’t even at 🤣 — momentum’s strong, keep it up team! www.vita-group.space/team

I really wanna see / do research on using AI to prevent massive shooting …

Perhaps I’m not alone in feeling GPT5 (even pro) is no better than o3-pro… coding and knowledge?

www.linkedin.com/posts/atlas-... Cannot resist enjoying some proud advisor moment

Eight years ago, I walked into my faculty office for the first time, not knowing what the journey ahead would hold. | Atlas Wang | 10 comments

Eight years ago, I walked into my faculty office for the first time, not knowing what the journey ahead would hold. This past year, while I’ve been gradually “fading into the background,” the group ...

linkedin.com

Two fresh tutorials on low-dimensional DL: Sparsity ’23 🔗 vita-group.github.io/SparsityTuto... Low-Rank ’25 🔗 vita-group.github.io/LowrankTutor... Curious about how sparsity and low-rank ideas power modern deep learning? I hope these slide decks & notes are a useful starting point! 🚀

Of all my students’ career achievements, this one makes me proudest: because really, who needs yet another AI professor or scientist alumni when you can boast a professional cat-breeder student? www.bestbritishcats.com To my Bay Area friends: BUY one! Maybe try using my name to get a discount :)

@talkachman.bsky.social caught it faster before I did... but here we go! 🚀 New pre-print dropped lnkd.in/gmQ9hBdf Its first draft won the #DARPA Disruptive Idea award at #NeuS2025. Now my student Peihao Wang has sharpened the theory even more. I’m so excited what it means for AI reasoning 🧩🤖 (1/n)

@talkachman.bsky.social · last yr.

@atlaswang.bsky.social knocking out of the park again! half a page in and this is a phenomenal read. So much to wrap my head around www.arxiv.org/abs/2506.21797

@atlaswang.bsky.social knocking out of the park again! half a page in and this is a phenomenal read. So much to wrap my head around www.arxiv.org/abs/2506.21797

Why Neural Network Can Discover Symbolic Structures with Gradient-based Training: An Algebraic and Geometric Foundation for Neurosymbolic Reasoning

We develop a theoretical framework that explains how discrete symbolic structures can emerge naturally from continuous neural network training dynamics. By lifting neural parameters to a measure space...

arxiv.org

Feels great relief to report: #Thrilledtoshare (yes, that LinkedIn tone I hate...) we have 0 papers accepted to @iccv.bsky.social ! Surprisingly liberating to step off the number game: I meant it. I’m genuinely proud of the science my students are building, no matter how ICCV reviewers think 🤣

An #AI system that automatically interprets echocardiograms maintained high accuracy across geography and time from complete and limited studies. https://ja.ma/4jYWBgI

Check out #PanEcho, now in @jama.com ✅ open weights ✅ multiview multitask reporting ✅ international validation ✅ works with POCUS Amazing effort @cardslab.bsky.social @giholste.bsky.social @rohankhera.bsky.social + @atlaswang.bsky.social, Márton Tokodi & Attila Kovács #CardioSky #EchoSky

JAMA@jama.com · last yr.

An #AI system that automatically interprets echocardiograms maintained high accuracy across geography and time from complete and limited studies. https://ja.ma/4jYWBgI

A reflection sparked by Saturday night’s thought-provoking dinner conversation 🥘 www.linkedin.com/pulse/beyond...

Beyond Bigger Models: AI Needs Bigger Problems

(Disclaimer: AI helped revise this article, too: -) “If AI is truly transformative, why are its dividends still so concentrated?” Over the past two years, I have watched two headlines run in parallel:...

linkedin.com

Making a new website of my research group, and did a visualization of all our papers from 2018 to present: clustered into 10 topics. One can clearly see how this group evolves its own tastes! … and deeper in my heart: long live optimization!! ❤️

Quoting one slides from @yann-lecun.bsky.social talk… arxiv is filled by papers that treat symptoms (or not even!) without ever diagnosing the disease

Unwelcome idea: I’m extremely tired of seeing one more LLM paper and believe academics shall pivot away from this area. So boring arxiv feeds every day…

This is going to kneecap science in this country for years. www.nature.com/articles/d41...

Exclusive: NSF stops awarding new grants and funding existing ones

US science funder also plans to screen grant applications for compliance with ‘agency priorities’.

nature.com

One of my PhD students got their visa revoked. I know of other cases amongst my AI colleagues. This is not what investing in US leadership in AI looks like. www.aljazeera.com/news/2025/4/...

US revokes nearly 1,500 student visas: Who are the targets?

Hundreds of students have had their visas cancelled and find themselves in limbo.

aljazeera.com

🚀 Thrilled to announce SPIN-Bench!🚀 We all love seeing how smart LLMs can be-solving complex math, crafting beautiful text, and coding effortlessly. But how well do they handle real-world strategic complexity, cooperation, & social negotiation? Can they play well when things get tricky? Not quite!

Just had a meal that gifted me two rare treasures: 1️⃣ Meeting someone infinitely wiser than me. 2️⃣ They weren’t cold or mean—just gently showed me where I could grow. "To learn truth at dawn, I’d die content by dusk." ✨ Humility tastes better with kindness. #Gratitude #LifeLessons

Good news? SPIN-Bench pinpoints exactly where these models fall short—and illuminates exciting research directions to smarter, socially savvy AI For some extra fun (and detailed, interactive trajectory visualizations!), visit project page： 👉 spinbench.github.io full paper: arxiv.org/pdf/2503.12349

🌍 Diplomacy – The ultimate test！ Models had to negotiate, forge alliances, and occasionally backstab. Result? Even the best LLMs floundered, struggling to juggle complex social interactions and strategic depth.

🎴 Cooperative Games (Hanabi): Coordination among teammates dramatically challenges models, causing their performance to dip sharply as complexity ramps up. Turns out, keeping track of your teammates’ intentions isn’t an easy task—even for GPTs!

Results fascinatingly reveal: 🔹 Classic Planning: LLMs ace simpler puzzles but struggle badly as complexity grows—losing track at longer-term decisions 🔎 Competitive Games: Top chess engines swept every LLM clean. Even simple tactical awareness quickly fades when facing deeper strategic branches