Subbarao Kambhampati (కంభంపాటి సుబ్బారావు)

Planning & Reasoning Abilities of LLMs/LRMs (Lecture 2 @ Melbourne ML Summer School 2026) | Subbarao Kambhampati 𝗣𝗼𝘀𝘁-𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗼𝗿 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝘀𝗲𝗹𝗳-𝗱𝗶𝘀𝘁𝗶𝗹𝗹𝗮𝘁𝗶𝗼𝗻 𝗶𝘀 𝗮𝗹𝗹 𝗮𝗯𝗼𝘂𝘁 𝗰𝗼𝗺𝗽𝗶𝗹𝗶𝗻𝗴 𝘃𝗲𝗿𝗶𝗳𝗶𝗲𝗿 𝘀𝗶𝗴𝗻𝗮𝗹 𝗶𝗻𝘁𝗼 𝘁𝗵𝗲 𝗯𝗮𝘀𝗲 𝗟𝗟𝗠 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗼𝗿 #SundayHarangue 𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱 𝗣𝗼𝘀𝘁-𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 ≈ 𝗖𝗼𝗺𝗽𝗶𝗹𝗶𝗻𝗴 𝗩𝗲𝗿𝗶𝗳𝗶𝗲𝗿 𝗦𝗶𝗴𝗻𝗮𝗹 𝗦𝗲𝗹𝗳-𝗱𝗶𝘀𝘁𝗶𝗹𝗹𝗮𝘁𝗶𝗼...

𝗣𝗼𝘀𝘁-𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴 𝘄𝗶𝘁𝗵 𝗼𝗿 𝘄𝗶𝘁𝗵𝗼𝘂𝘁 𝘀𝗲𝗹𝗳-𝗱𝗶𝘀𝘁𝗶𝗹𝗹𝗮𝘁𝗶𝗼𝗻 𝗶𝘀 𝗮𝗹𝗹 𝗮𝗯𝗼𝘂𝘁 𝗰𝗼𝗺𝗽𝗶𝗹𝗶𝗻𝗴 𝘃𝗲𝗿𝗶𝗳𝗶𝗲𝗿 𝘀𝗶𝗴𝗻𝗮𝗹 𝗶𝗻𝘁𝗼 𝘁𝗵𝗲 𝗯𝗮𝘀𝗲 𝗟𝗟𝗠 𝗴𝗲𝗻𝗲𝗿𝗮𝘁𝗼𝗿 #SundayHarangue

𝗦𝘁𝗮𝗻𝗱𝗮𝗿𝗱 𝗣𝗼𝘀𝘁-𝗧𝗿𝗮𝗶𝗻𝗶𝗻𝗴 ≈ 𝗖𝗼𝗺𝗽𝗶𝗹𝗶𝗻𝗴 𝗩𝗲𝗿𝗶𝗳𝗶𝗲𝗿 𝗦𝗶𝗴𝗻𝗮𝗹
𝗦𝗲𝗹𝗳-𝗱𝗶𝘀𝘁𝗶𝗹𝗹𝗮𝘁𝗶𝗼𝗻 = 𝗖𝗼𝗺𝗽𝗶𝗹𝗶𝗻𝗴 𝗩𝗲𝗿𝗶𝗳𝗶𝗲𝗿 𝗙𝗲𝗲𝗱𝗯𝗮𝗰𝗸 𝘁𝗼𝗼

See 👇 for more..

www.linkedin.com/posts/subbar...

Does it really make sense to think of inference efficiency in terms of the number of tokens produced?

No. 👇

x.com/i/status/202...

Sorry, but I think you miss the point that most of the reasoning model revolution came exactly for tasks where there are verifiers--whether external/symbolic, or learned, or even hand-coded simulators. What do you think RLVR or Self Distillation are?

The lectures, 3hrs long with Q&A, are quite up-to-date and cover LLMs, LRMs, as well as the latest test-time scaling and post-training methods such as LLM-Process-Modulo and self-distillation.

Here are the recordings of two lectures on 𝗣𝗹𝗮𝗻𝗻𝗶𝗻𝗴 & 𝗥𝗲𝗮𝘀𝗼𝗻𝗶𝗻𝗴 𝗖𝗮𝗽𝗮𝗯𝗶𝗹𝗶𝘁𝗶𝗲𝘀 𝗼𝗳 𝗟𝗟𝗠𝘀/𝗟𝗥𝗠𝘀 that I gave this week at Melbourne ML Summer School (lnkd.in/g7rxg9sw).

𝙇𝙚𝙘𝙩𝙪𝙧𝙚 1: youtube.com/watch?v=_PPV...
𝙇𝙚𝙘𝙩𝙪𝙧𝙚 2: youtube.com/watch?v=fKlm...

Slides available with the video (direct link bit.ly/4sXyjtj)

YouTube video by Subbarao Kambhampati Anthropomorphization Sins in Modern AI (or Perils of Prematurely Applying Lens of Cognition to LLMs)

A common theme in our work these past few years has been pushing back on facile anthropomorphizations (and/or efforts that bring questionable/discredited Cognitive Science metaphors) to LLMs.. So I enjoyed giving this talk at @ivado.bsky.social yesterday... www.youtube.com/watch?v=CoyS...

YouTube video by Subbarao Kambhampati On the Mythos of LRM "Thinking Tokens" (Talk @ Microsoft Research, India; 12/16/2025)

Three of my talks in India last month--at @iitdelhi.bsky.social,
@msftresearch.bsky.social India and at IndoML Symposium--were "On the Mythos of LRM Thinking Tokens." Here is a recording of one of them--the talk I gave at MSR India.

www.youtube.com/watch?v=fCQX...

Like I say, if a human--even a Terence Tao--makes an egregious mistake (e.g. the one below) once, our trust in them takes a nose dive. With LLMs, it is just "..but they do so well on IMO problems!"..

YouTube video by Subbarao Kambhampati Talk on the semantics of "Thinking Traces" (Keynote at NeurIPS2025 MAR Workshop)

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 the seven papers we presented at the conference workshops. Special thanks to the engaged audience..🙏

www.youtube.com/watch?v=rvby...

[On using Continuous Latent Space Vectors in the context windows of Transformers and LLMs] #SundayHarangue
👉 x.com/rao2z/status...

YouTube video by Subbarao Kambhampati LRMs and Agentic AI (Talk at Samsung AI Forum)

My talk at Samsung AI Forum yesterday
www.youtube.com/watch?v=L2nA...

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… | Subbarao K... 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..) 𝗘𝘃𝗮𝗹...

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..👇

www.linkedin.com/posts/subbar...

𝐏𝐞𝐫𝐟𝐨𝐫𝐦𝐚𝐭𝐢𝐯𝐞 𝐓𝐡𝐢𝐧𝐤𝐢𝐧𝐠? The anthropomorphization of LRM intermediate tokens as thinking begat a cottage industry to "get efficiency by shortening thinking." We ask: 𝗜𝘀 𝗖𝗼𝗧 𝗹𝗲𝗻𝗴𝘁𝗵 𝗿𝗲𝗮𝗹𝗹𝘆 𝗮 𝗿𝗲𝗳𝗹𝗲𝗰𝘁𝗶𝗼𝗻 𝗼𝗳 𝗽𝗿𝗼𝗯𝗹𝗲𝗺 𝗵𝗮𝗿𝗱𝗻𝗲𝘀𝘀 𝗼𝗿 𝗶𝘀 𝗶𝘁 𝗺𝗼𝗿𝗲 𝗽𝗲𝗿𝗳𝗼𝗿𝗺𝗮𝘁𝗶𝘃𝗲? 👉 www.linkedin.com/posts/subbar...

Subbarao Kambhampati (కంభంపాటి సుబ్బారావు) on X: "Rejecting papers in #AI Conferences because of "resource constraints" is shooting ourselves in the foot as a community; use Findings.. #SundayHarangue By now, we have all know that top AI conferences are oversubscribed (in terms of paper submissions), and have heard that that" / X Rejecting papers in #AI Conferences because of "resource constraints" is shooting ourselves in the foot as a community; use Findings.. #SundayHarangue By now, we have all know that top AI conferences are oversubscribed (in terms of paper submissions), and have heard that that

Rejecting papers in #AI Conferences because of "resource constraints" is shooting ourselves in the foot as a community; use Findings.. #SundayHarangue 👇

x.com/rao2z/status...

Subbarao Kambhampati (కంభంపాటి సుబ్బారావు) on X: "Proofs are not reasoning traces & I/O Format Language shouldn't be much of an issue for LLMs #SundayHarangue (Special IMO edition). 1/ My feed these last couple of days of IMO discussions has been full of comments that seem to conflate LRM intermediate tokens (aka reasoning" / X Proofs are not reasoning traces & I/O Format Language shouldn't be much of an issue for LLMs #SundayHarangue (Special IMO edition). 1/ My feed these last couple of days of IMO discussions has been full of comments that seem to conflate LRM intermediate tokens (aka reasoning

Proofs are not reasoning traces & I/O Format Language shouldn't be much of an issue for LLMs + other things #SundayHarangue (Special IMO edition). 🧵 👇

x.com/rao2z/status...

Neither LLMs nor LRMs have the ability to go beyond the humanity's knowledge closure--which is needed for true discoveries. | Subbarao Kambhampati Neither LLMs nor LRMs have the ability to go beyond the humanity's knowledge closure--which is needed for true discoveries. Both are beholden to the collected knowledge of the humanity (whether de...

Both LLMs and LRMs are upper bounded by humanity's knowledge closure. True scientific discoveries are, by definition, outside of that closure. Ergo, LLMs/LRMs are great force multipliers to us; but don't support "Nobel this weekend" hype..

👉 www.linkedin.com/posts/subbar...

Computational Complexity is the wrong measure for LRMs (as it was for LLMs)--think distributional distance instead #SundayHarangue (yes, we're back!)

👉 x.com/rao2z/status...

A̶̶̶I̶̶̶ ̶ ̶ ̶ ̶(̶A̶r̶t̶i̶f̶i̶c̶i̶a̶l̶ ̶I̶n̶t̶e̶l̶l̶i̶g̶e̶n̶c̶e̶)̶
̶̶̶A̶̶̶G̶̶̶I̶̶̶ ̶(̶A̶r̶t̶i̶f̶i̶c̶i̶a̶l̶ ̶G̶e̶n̶e̶r̶a̶l̶ ̶I̶n̶t̶e̶l̶l̶i̶g̶e̶n̶c̶e̶)̶
̶̶̶A̶̶̶S̶̶̶I̶̶̶ ̶(̶A̶r̶t̶i̶f̶i̶c̶i̶a̶l̶ ̶S̶u̶p̶e̶r̶ ̶I̶n̶t̶e̶l̶l̶i̶g̶e̶n̶c̶e̶)
ASDI (Artificial Super Duper Intelligence)

Don't get stuck with yesterday's hypeonyms!
Dare to get to the next level!

#AIAphorisms

Subbarao Kambhampati (కంభంపాటి సుబ్బారావు) on X: "Some of what that recent Apple LRM limitations paper shows is known (pardon my friendly Schmidhubering; I do welcome more LLM studies with scientific skepticism). Our study 👇 from Sep 2024 shows o1 accuracy degrading as complexity increases.. 1/ https://t.co/d8zEUGi4SZ" / X Some of what that recent Apple LRM limitations paper shows is known (pardon my friendly Schmidhubering; I do welcome more LLM studies with scientific skepticism). Our study 👇 from Sep 2024 shows o1 accuracy degrading as complexity increases.. 1/ https://t.co/d8zEUGi4SZ

This series of lectures was given the same week there was all that brouhaha over the Apple illusion paper (I was giving these lectures during the day and talking to reporters in the evening 😅). As such they are pretty up-to-date! 3/

x.com/rao2z/status...

The lectures start with a "big picture" overview (Lecture 1); focus on standard LLMs and their limitations, and LLM-Modulo as a test-time scaling approach (Lecture 2); and end with a critical appraisal of the test-time scaling and RL post-training techniques (Lecture 3). 2/

ACDL Summer School Lectures on Planning/Reasoning Abilities of LLMs/LRMs - YouTube

For anyone interested, here are the videos of the three ~50min each lectures on the reasoning/planning capabilities of LLMs/LRMs that I gave at #ACDL2025 in Riva Del Sole resort last week. 1/

www.youtube.com/playlist?lis...

...it basically confirmed what is already well-established: LLMs (& LRMs & "LLM agents") have trouble w/ problems that require many steps of reasoning/planning.

See, e.g., lots of recent papers by Subbarao Kambhampati's group at ASU. (2/2)

An AGI-wannabe reasoning model whining that it couldn't handle a problem because its context window isn't big enough is like a superman-wannabe little kid protesting that he couldn't add those numbers because he doesn't have enough fingers and toes.. #AIAphorisms

Beyond Semantics: The Unreasonable Effectiveness of Reasonless Intermediate Tokens Recent impressive results from large reasoning models have been interpreted as a triumph of Chain of Thought (CoT), and especially of the process of training on CoTs sampled from base LLMs in order to...

"our counter-intuitive results demonstrate ways in which common interpretations of Large Reasoning Models may be anthropomorphizations or simplifications" arxiv.org/abs/2505.13775

Lucas Saldyt on X: "Neural networks can express more than they learn, creating expressivity-trainability gaps. Our paper, “Mind The Gap,” shows neural networks best learn parallel algorithms, and analyzes gaps in faithfulness and effectiveness. @rao2z https://t.co/8YjxPkXFu0" / X Neural networks can express more than they learn, creating expressivity-trainability gaps. Our paper, “Mind The Gap,” shows neural networks best learn parallel algorithms, and analyzes gaps in faithfulness and effectiveness. @rao2z https://t.co/8YjxPkXFu0

The transformer expressiveness results are often a bit of a red herring as there tends to be a huge gap between what can be expressed in transformers, and what can be learned with gradient descent. Mind the Gap, a new paper with
Lucas Saldyt dives deeper into this issue 👇👇

x.com/SaldytLucas/...

Anthropomorphization of intermediate tokens as reasoning/thinking traces isn't quite a harmless fad, and may be pushing LRM research into questionable directions.. So we decided to put together a more complete argument. Paper 👉 arxiv.org/pdf/2504.09762 (Twitter thread: x.com/rao2z/status...)

Subbarao Kambhampati (కంభంపాటి సుబ్బారావు) on X: "This RLiNo? paper lead by @soumya_samineni & @durgesh_kalwar dives into the MDP model used in the RL post-training methods inspired by DeepSeek R1, and asks if some of the idiosyncrasies of RL in R1 aren't just consequences of the simplistic structural assumptions in the MDP 🧵1/ https://t.co/qPBY3tJILE" / X This RLiNo? paper lead by @soumya_samineni & @durgesh_kalwar dives into the MDP model used in the RL post-training methods inspired by DeepSeek R1, and asks if some of the idiosyncrasies of RL in R1 aren't just consequences of the simplistic structural assumptions in the MDP 🧵1/ https://t.co/qPBY3tJILE

Longer thread here

x.com/rao2z/status...

This RLiNo? paper (arxiv.org/abs/2505.13697) lead by Soumya Samineni and Durgesh_kalwar dives into the MDP model used in the RL post-training methods inspired by DeepSeek R1, and asks if some of the idiosyncrasies of RL aren't just consequences of the simplistic structural assumptions made

Do Intermediate Tokens Produced by LRMs (need to) have any semantics? Our new study 👇

Thread 👉 x.com/rao2z/status...