Subbarao Kambhampati (కంభంపాటి సుబ్బారావు)
@rao2z
Followers
27K
Following
4K
Media
4K
Statuses
11K
AI researcher & teacher @SCAI_ASU. Former President of @RealAAAI; Chair of @AAAS Sec T. Here to tweach #AI. YouTube Ch: https://t.co/4beUPOmf6y Bsky: rao2z
Tempe, AZ
Joined October 2014
In the year since LRMs ("reasoning models") hit the scene, we have been trying to understand, analyze and demystify them.. Here are our efforts to date--conveniently all in one place (First..) Evaluation of LRMs on Planning 📜 https://t.co/Y527iqlJHW (9/24)&
10
64
312
You can tell that a talk is really good if you're simultaneously thinking "this is obvious" and "why did I not realize this before?"
ICYMI, here is my keynote on the semantics of LRM "thinking traces" at #NeurIPS2025 workshop on Multimodal Algorithmic Reasoning. It's a unified view of seven papers we presented at the conference workshops. Special thanks to the attentive audience🙏 👉 https://t.co/PYLd97vBKQ
1
2
5
ICYMI, here is my keynote on the semantics of LRM "thinking traces" at #NeurIPS2025 workshop on Multimodal Algorithmic Reasoning. It's a unified view of seven papers we presented at the conference workshops. Special thanks to the attentive audience🙏 👉 https://t.co/PYLd97vBKQ
2
11
45
@pmddomingos Thanks for the shoutout! 🙏 Here is the poster on that paper as presented at #NeurIPS2025 LAW & ForLM workshops yesterday by @karthikv792, @PalodVardh12428 & @kayastechly
0
2
8
Shocking but true: LLMs learn to reason better when you give them the wrong traces.
arxiv.org
Recent impressive results from large reasoning models have been interpreted as a triumph of Chain of Thought (CoT), especially of training on CoTs sampled from base LLMs to help find new reasoning...
12
41
324
Behold the service that charges $3,325 to help high school students submit applied AI papers by the dozens. The @neuripsconf initiatives for highschool student papers etc are not blameless in creating incentives for these papermills..IMHO. #neurips2025
https://t.co/LxduOVDMRj
theguardian.com
AI research in question as author claims to have written over 100 papers on AI that one expert calls a ‘disaster’
0
1
15
You gotta celebrate surviving giving a spicy talk with equally spicy food, no? 😅#NeurIPS2025
📢I will be speaking at the #NeurIPS2025 Multimodal Algorithmic Reasoning workshop tomorrow about the mythos of thinking tokens. (Sunday 7th at 11:30AM; Upper Level 11AB; https://t.co/2RuyFcw6aK). It will be an overview of several lines of work Yochanites are presenting at
3
0
52
At the #NeurIPS2025 Efficient Reasoning workshop yesterday, apparently someone exclaimed to my students presenting this 👇work: "But that seems to contradict the other papers here." Well, sometimes, the emperor is wearing diaphanous clothes.. 😅
The anthropomorphization of LRM intermediate tokens as thinking begat a cottage industry to "get efficiency by shortening thinking." @PalodVardh12428 (w/ @karthikv792 & @kayastechly) asks: Is CoT length really a reflection of problem hardness or is it more performative? 1/
1
5
18
This 👇happens at 11:30 in 11AB.. 🤞 #NeurIPS2025 The time has come, the walrus said To talk of many things Of Think Tokens, Semantics, Complexity and whether R1 RL even has any RL.. https://t.co/bRBScFmSHC
📢I will be speaking at the #NeurIPS2025 Multimodal Algorithmic Reasoning workshop tomorrow about the mythos of thinking tokens. (Sunday 7th at 11:30AM; Upper Level 11AB; https://t.co/2RuyFcw6aK). It will be an overview of several lines of work Yochanites are presenting at
0
3
12
Swing by! Live now at NeurIPS 2025 → the Multimodal Algorithmic Reasoning Workshop!📍 Upper Level, Room 11A/B This morning’s powerhouse lineup: • Noah Goodman @noahdgoodman (9:15–9:45): Learning to Reason • Lindsey Li @LINJIEFUN (9:45–10:15): Pictures think harder than words
0
4
6
(The rest) https://t.co/TssClsh1aX
#NeurIPS2025 is going to end with a BANG for Yochanites.. We have multiple papers being presented at LAW (4), CogInterp (3) & ForLM (2) workshops. I will also be giving a keynote at MAR workshop (11:30AM 11AB). Wish us luck..😬 🧵 0/
0
0
2
4/ MAR https://t.co/BPYHVexSFs
📢I will be speaking at the #NeurIPS2025 Multimodal Algorithmic Reasoning workshop tomorrow about the mythos of thinking tokens. (Sunday 7th at 11:30AM; Upper Level 11AB; https://t.co/2RuyFcw6aK). It will be an overview of several lines of work Yochanites are presenting at
0
1
3
3/ ForLM https://t.co/Z8amdtnl1w
📢 Yochanites are presenting two papers at the #NeurIPS2025 ForLM workshop today (Ballroom 33ABC; https://t.co/VBvwdj8m0i): 👉Beyond Semantics: The Unreasonable Effectiveness of Reasonless Intermediate Tokens 👉RL in Name Only? Analyzing the Structural Assumptions in RL
1
1
3
2/ CogInterp https://t.co/Gc4mZIQPAd
📢 Yochanites are presenting three papers at #NeurIPS CogInterp Workshop today (Ballroom 5AB; https://t.co/fSj5XXLW5X) 👉Stop Anthropomorphizing Intermediate Toakens as Reasoning/Thinking Traces 👉Do Cognitively Interpretable Reasoning Traces Improve LLM Performance?
1
1
2
1/ LAW https://t.co/qSiIf1Ccu8
📢Yochanites are presenting four papers at the #NeurIPS2025 LAW 2025 workshop (Ballroom 20D; https://t.co/ZfySCwkNQm) today--on the semantics (or lack there of) of intermediate tokens, on whether the DeepSeek R1 RL is "RL in name only?" and a position arguing against the
1
1
2
#NeurIPS2025 is going to end with a BANG for Yochanites.. We have multiple papers being presented at LAW (4), CogInterp (3) & ForLM (2) workshops. I will also be giving a keynote at MAR workshop (11:30AM 11AB). Wish us luck..😬 🧵 0/
2
2
11
📢 Yochanites are presenting two papers at the #NeurIPS2025 ForLM workshop today (Ballroom 33ABC; https://t.co/VBvwdj8m0i): 👉Beyond Semantics: The Unreasonable Effectiveness of Reasonless Intermediate Tokens 👉RL in Name Only? Analyzing the Structural Assumptions in RL
0
3
13
📢 Yochanites are presenting three papers at #NeurIPS CogInterp Workshop today (Ballroom 5AB; https://t.co/fSj5XXLW5X) 👉Stop Anthropomorphizing Intermediate Toakens as Reasoning/Thinking Traces 👉Do Cognitively Interpretable Reasoning Traces Improve LLM Performance?
0
6
13