Uploaded February 2025 | Updated November 2026, 4 days ago
SPONSOR MESSAGES:
***
CentML offers competitive pricing for GenAI model deployment, with flexible options to suit a wide range of models, from small to large-scale deployments. Check out their super fast DeepSeek R1 hosting!
centml.ai/pricing
Sepp Hochreiter, the inventor of LSTM networks, makes a forceful case that large language models are fundamentally limited -- they store and retrieve human knowledge rather than reason about it. He walks through the technical history from the original 1991 vanishing gradient problem through to his latest work on xLSTM, explaining how exponential gating and matrix memory replace the old sigmoid bottleneck to create architectures that can compete with Transformers on sequence modeling while maintaining fixed memory footprints.
The conversation covers the practical advantages of xLSTM for industrial applications like robotics simulation and discrete element modeling, where constant memory consumption matters more than raw benchmark performance. Hochreiter also discusses the neuro-symbolic Pi AI project at JKU Linz, his view that genuine reasoning requires something beyond pattern matching on training data, and why he believes the current scaling paradigm will hit a wall. The technical deep-dive into FlashAttention comparisons, recurrent vs attention-based architectures, and the evolution from sigmoid to exponential gating mechanisms gives a clear picture of where LSTM-style architectures fit in the current landscape.
---
REFERENCES:
Paper:
[00:00:13] Long Short-Term Memory (Original LSTM Paper)
https://direct.mit.edu/neco/article-abstract/9/8/1735/6109/Long-Short-Term-Memory
[00:04:18] Kolmogorov Complexity
link.springer.com/article/10.1007/BF02478259
[00:19:38] xLSTM: Extended Long Short-Term Memory
arxiv.org/abs/2405.04517
[00:22:53] FlashAttention: Fast and Memory-Efficient Attention
arxiv.org/abs/2205.14135
[00:36:08] Mamba: Linear-Time Sequence Modeling
arxiv.org/abs/2312.00752
[00:55:03] Core Knowledge Theory of Human Cognition
harvardlds.org/wp-content/uploads/2017/01/SpelkeKinzler07-1.pdf
Book:
[00:53:18] Thinking, Fast and Slow
amazon.com/Thinking-Fast-Slow-Daniel-Kahneman/dp/0374533555
Company:
[01:00:18] NXAI: Industrial AI with xLSTM
https://investinaustria.at/en/blog/nxai-top-researchers-in-austria-develop-solutions-for-industrial-companies/
---
LINKS:
Full Transcript: app.rescript.info/share/50e3f79d116b1a92aa40819b6cb40eb2
Download PDF transcript: app.rescript.info/api/public/sessions/bdb8e291a0e4514d/pdf
Prof. Sepp Hochreiter
nx-ai.com
https://x.com/hochreitersepp
https://scholar.google.at/citations?user=tvUH3WMAAAAJ&hl=en
SPONSOR MESSAGES:
***
CentML offers competitive pricing for GenAI model deployment, with flexible options to suit a wide range of models, from small to large-scale deployments. Check out their super fast DeepSeek R1 hosting!
centml.ai/pricing
Sepp Hochreiter, the inventor of LSTM networks, makes a forceful case that large language models are fundamentally limited -- they store and retrieve human knowledge rather than reason about it. He walks through the technical history from the original 1991 vanishing gradient problem through to his latest work on xLSTM, explaining how exponential gating and matrix memory replace the old sigmoid bottleneck to create architectures that can compete with Transformers on sequence modeling while maintaining fixed memory footprints.
The conversation covers the practical advantages of xLSTM for industrial applications like robotics simulation and discrete element modeling, where constant memory consumption matters more than raw benchmark performance. Hochreiter also discusses the neuro-symbolic Pi AI project at JKU Linz, his view that genuine reasoning requires something beyond pattern matching on training data, and why he believes the current scaling paradigm will hit a wall. The technical deep-dive into FlashAttention comparisons, recurrent vs attention-based architectures, and the evolution from sigmoid to exponential gating mechanisms gives a clear picture of where LSTM-style architectures fit in the current landscape.
---
REFERENCES:
Paper:
[00:00:13] Long Short-Term Memory (Original LSTM Paper)
https://direct.mit.edu/neco/article-abstract/9/8/1735/6109/Long-Short-Term-Memory
[00:04:18] Kolmogorov Complexity
link.springer.com/article/10.1007/BF02478259
[00:19:38] xLSTM: Extended Long Short-Term Memory
arxiv.org/abs/2405.04517
[00:22:53] FlashAttention: Fast and Memory-Efficient Attention
arxiv.org/abs/2205.14135
[00:36:08] Mamba: Linear-Time Sequence Modeling
arxiv.org/abs/2312.00752
[00:55:03] Core Knowledge Theory of Human Cognition
harvardlds.org/wp-content/uploads/2017/01/SpelkeKinzler07-1.pdf
Book:
[00:53:18] Thinking, Fast and Slow
amazon.com/Thinking-Fast-Slow-Daniel-Kahneman/dp/0374533555
Company:
[01:00:18] NXAI: Industrial AI with xLSTM
https://investinaustria.at/en/blog/nxai-top-researchers-in-austria-develop-solutions-for-industrial-companies/
---
LINKS:
Full Transcript: app.rescript.info/share/50e3f79d116b1a92aa40819b6cb40eb2
Download PDF transcript: app.rescript.info/api/public/sessions/bdb8e291a0e4514d/pdf
Prof. Sepp Hochreiter
nx-ai.com
https://x.com/hochreitersepp
https://scholar.google.at/citations?user=tvUH3WMAAAAJ&hl=en
![New Research: LMArena is rigged
We expose the serious problems with Chatbot Arena, the AI industrys most influential leaderboard. We showcase the recent paper The Leaderboard Illusion which shows that its actually being gamed by the biggest players in tech.
We comment on the recent revelation that Mark Zuckerberg openly admitted that Meta manipulated the arena by fine-tuning models specifically to top the charts. The paper uncovers how select companies receive preferential treatment, including extensive private testing and the ability to retract underperforming models. These companies also get a massive data advantage, as their models are sampled far more often, allowing them to train on arena data to boost their scores.
[HEDGING]
Metas actions were actually useful to expose a serious flaw in the arena and Zuck 100% owned it, and was super transparent about it afterwards.
SPONSOR MESSAGES:
Prolific: Quality data. From real people. For faster breakthroughs.
https://prolific.com/mlst?utm_campaign=98404559-MLST&utm_source=youtube&utm_medium=podcast&utm_content=script-4
Tufa AI Labs are hiring for ML Engineers and a Chief Scientist in Zurich/SF
https://tufalabs.ai/
Interactive Transcript:
https://app.rescript.info/public/share/I8k9saMwSQbkpL7rbZRiWCKnFHzgSLBi2atvnVACVm8
TOC:
00:00:00 - The Serious Problem with Chatbot Arena
00:01:45 - Mark Zuckerberg Admits to Hacking the Leaderboard
00:03:26 - Goodharts Law: Why the Benchmark is Broken
00:04:00 - Andrej Karpathy on Judging Subtle Model Improvements
00:06:45 - The Origin Story of Chatbot Arena
00:10:00 - How ELO Ratings Work (and Their Flaws)
00:15:15 - The Explosive Leaderboard Illusion Paper by Cohere
00:16:45 - Finding #1: Preferential Private Testing
00:18:00 - Finding #2: Unfair Data Advantage for Big Tech
00:19:15 - Finding #3: Massive Gains from Fine-Tuning on Arena Data
00:20:45 - Why Some Models Are Sampled Overwhelmingly More Than Others
00:22:15 - Coheres Proposed Solutions to Fix the Arena
00:23:45 - Analysis: How Repetitive Are User Prompts?
00:25:56 - Final Thoughts & The Arenas Controversial Response
REFS:
The Leaderboard Illusion [2025]
Shivalika Singh, Yiyang Nan, Alex Wang, Daniel DSouza, Sayash Kapoor, Ahmet Üstün, Sanmi Koyejo, Yuntian Deng, Shayne Longpre, Noah A. Smith, Beyza Ermis, Marzieh Fadaee, Sara Hooker
https://arxiv.org/abs/2504.20879
LM Arena response:
https://blog.lmarena.ai/blog/2025/our-response/#:~:text=Published&text=Recently%2C%20a%20writeup%20titled%20%E2%80%9CThe,ongoing%20discussions%20with%20the%20authors.
Elo Uncovered: Robustness and Best Practices in Language Model Evaluation [2023]
https://arxiv.org/pdf/2311.17295
Meriem Boubdir, Edward Kim, Beyza Ermis, Sara Hooker, Marzieh Fadaee
Chatbot Arena: An Open Platform for Evaluating LLMs by Human Preference [2024]
https://arxiv.org/abs/2403.04132
Wei-Lin Chiang, Lianmin Zheng, Ying Sheng, Anastasios Nikolas Angelopoulos, Tianle Li, Dacheng Li, Hao Zhang, Banghua Zhu, Michael Jordan, Joseph E. Gonzalez, Ion Stoica
Too much efficiency makes everything worse: overfitting and the strong version of Goodharts law
https://sohl-dickstein.github.io/2022/11/06/strong-Goodhart.html
Jascha Sohl-Dickstein
Primary/Senior Authors
https://x.com/singhshiviii
https://x.com/mziizm
https://x.com/sarahookr New Research: LMArena is rigged](https://i.ytimg.com/vi/9W_OhS38rIE/mqdefault.jpg)
![AutoGrad Changed Everything (Not Transformers) [Dr. Jeff Beck]
Dr. Jeff Beck, mathematician turned computational neuroscientist, joins us for a fascinating deep dive into why the future of AI might look less like ChatGPT and more like your own brain.
**SPONSOR MESSAGES START**
—
Prolific - Quality data. From real people. For faster breakthroughs.
https://www.prolific.com/?utm_source=mlst
—
**END**
*What if the key to building truly intelligent machines isnt bigger models, but smarter ones?*
In this conversation, Jeff makes a compelling case that weve been building AI backwards. While the tech industry races to scale up transformers and language models, Jeff argues were missing something fundamental: the brain doesnt work like a giant prediction engine. It works like a scientist, constantly testing hypotheses about a world made of *objects* that interact through *forces* — not pixels and tokens.
*The Bayesian Brain* — Jeff explains how your brain is essentially running the scientific method on autopilot. When you combine what you see with what you hear, youre doing optimal Bayesian inference without even knowing it. This isnt just philosophy — its backed by decades of behavioral experiments showing humans are surprisingly efficient at handling uncertainty.
*AutoGrad Changed Everything* — Forget transformers for a moment. Jeff argues the real hero of the AI boom was automatic differentiation, which turned AI from a math problem into an engineering problem. But in the process, we lost sight of what actually makes intelligence work.
*The Cat in the Warehouse Problem* — Heres where it gets practical. Imagine a warehouse robot thats never seen a cat. Current AI would either crash or make something up. Jeffs approach? Build models that *know what they dont know*, can phone a friend to download new object models on the fly, and keep learning continuously. Its like giving robots the ability to say wait, what IS that? instead of confidently being wrong.
*Why Language is a Terrible Model for Thought* — In a provocative twist, Jeff argues that grounding AI in language (like we do with LLMs) is fundamentally misguided. Self-report is the least reliable data in psychology — people routinely explain their own behavior incorrectly. We should be grounding AI in physics, not words.
*The Future is Lots of Little Models* — Instead of one massive neural network, Jeff envisions AI systems built like video game engines: thousands of small, modular object models that can be combined, swapped, and updated independently. Its more efficient, more flexible, and much closer to how we actually think.
Whether youre an AI researcher, a robotics enthusiast, or just curious about how minds — biological or artificial — actually work, this conversation offers a refreshingly different perspective on where intelligence comes from and where its going.
Rescript: https://app.rescript.info/public/share/D-b494t8DIV-KRGYONJghvg-aelMmxSDjKthjGdYqsE
TIMESTAMPS:
00:00:00 Introduction & The Bayesian Brain
00:01:25 Bayesian Inference & Information Processing
00:05:17 The Brain Metaphor: From Levers to Computers
00:10:13 Micro vs. Macro Causation & Instrumentalism
00:16:59 The Active Inference Community & AutoGrad
00:22:54 Object-Centered Models & The Grounding Problem
00:35:50 Scaling Bayesian Inference & Architecture Design
00:48:05 The Cat in the Warehouse: Solving Generalization
00:58:17 Alignment via Belief Exchange
01:05:24 Deception, Emergence & Cellular Automata
REFERENCES:
Paper:
[00:00:24] Zoubin Ghahramani (Google DeepMind)
https://pmc.ncbi.nlm.nih.gov/articles/PMC3538441/pdf/rsta201
[00:19:20] Mamba: Linear-Time Sequence Modeling
https://arxiv.org/abs/2312.00752
[00:27:36] xLSTM: Extended Long Short-Term Memory
https://arxiv.org/abs/2405.04517
[00:41:12] 3D Gaussian Splatting
https://repo-sam.inria.fr/fungraph/3d-gaussian-splatting/
[01:07:09] Lenia: Biology of Artificial Life
https://arxiv.org/abs/1812.05433
[01:08:20] Growing Neural Cellular Automata
https://distill.pub/2020/growing-ca/
[01:14:05] DreamCoder
https://arxiv.org/abs/2006.08381
[01:14:58] The Genomic Bottleneck
https://www.nature.com/articles/s41467-019-11786-6
Person:
[00:16:42] Karl Friston (UCL)
https://www.youtube.com/watch?v=PNYWi996Beg AutoGrad Changed Everything (Not Transformers) [Dr. Jeff Beck]](https://i.ytimg.com/vi/9suqiofCiwM/mqdefault.jpg)
![Intelligence is collective, not artificial — Prof. Michael I. Jordan (UC Berkeley / Inria)
Michael I. Jordan, described by Science magazine as the most influential computer scientist alive, has never thought of himself as an AI researcher. In this conversation he explains why that distinction matters.
SPONSOR:
Cyber Fund built the Monastery to help founders ship products that were impossible a year ago. Applications for Batch 1 are now open.
Apply now: https://cyber.fund
Jordan trained as a statistician and cognitive scientist, and his career has been spent building machine learning systems that work in the real world: supply chains, commerce, healthcare, and large economic systems. When the field rebranded itself as AI and then AGI, he did not follow. Instead he argues that the framing is wrong. AI is better understood as a collective economic system than as a race to build a disembodied superintelligence.
We talk about why AGI is mostly a PR term, what machine learning achieved before the LLM hype cycle, and why the assistant-on-your-shoulder vision may be less compelling than it sounds. Jordan explains why explanations need to be actionable, not merely mechanistic; why AlphaFolds missing error bars matter; how prediction-powered inference changes the picture; and why drug discovery is an incentive-design problem rather than a pure pattern-matching problem.
ERRATA: Science magazine ranked him the most influential computer scientist, not Nature
TIMESTAMPS:
00:00:00 Cold open: A demoralizing message to young builders
00:02:04 CyberFund sponsor read
00:02:50 From symbolic AI to machine learning systems
00:05:42 Why AGI is mostly a PR term
00:08:48 A collectivist, economic perspective on AI
00:11:33 Why LLMs need system design, not hype
00:14:50 Predictability beats faux understanding
00:17:55 AlphaFold, bias, and prediction-powered inference
00:21:48 Stop anthropomorphizing intelligence
00:27:44 Drug discovery as an incentive problem
00:32:29 The three-layer data market
00:38:07 Social knowledge, markets, and culture
00:45:39 Creator economics beyond Spotify
00:48:30 How science-fiction AI narratives mislead young builders
00:51:45 AI should improve humans, not replace them
00:56:42 Safety is a property of the whole system
00:58:12 Silicon Valley gurus and the cream off the top
01:00:47 Game theory, mechanism design, and contracts
01:04:39 Conformal prediction, e-values, and anytime inference
01:08:11 A new liberal arts triangle for the AI era
01:11:30 The Bayesian duck and markets as uncertainty reduction
REFERENCES:
person:
[00:02:50] Michael I. Jordan (homepage)
https://people.eecs.berkeley.edu/~jordan/
paper:
[00:06:01] A Collectivist, Economic Perspective on AI
https://arxiv.org/abs/2507.06268
[00:18:09] AlphaFold
https://www.nature.com/articles/s41586-021-03819-2
[00:20:36] Prediction-Powered Inference
https://arxiv.org/abs/2301.09633
[00:24:38] On the Measure of Intelligence
https://arxiv.org/abs/1911.01547
[00:33:47] On Three-Layer Data Markets
https://arxiv.org/abs/2402.09697
[01:04:39] Conformal Prediction with Conditional Guarantees
https://arxiv.org/abs/2107.07511
[01:04:51] A Tutorial on Conformal Prediction
https://www.jmlr.org/papers/v9/shafer08a.html
[01:06:00] E-Values Expand the Scope of Conformal Prediction
https://arxiv.org/abs/2503.13050
[01:08:23] Computational Thinking
https://www.cs.cmu.edu/~CompThink/papers/Wing06.pdf
other:
[00:11:33] The Bitter Lesson
http://www.incompleteideas.net/IncIdeas/BitterLesson.html
[00:20:50] How to use AI for discovery without leading science astray
https://news.berkeley.edu/2023/11/09/how-to-use-ai-for-discovery-without-leading-science-astray/
[00:28:20] How Should the FDA Test?
https://rdi.berkeley.edu/events/sbc-assets/pdfs/Summit%20session%20speaker%20slides%20submission%20form-s1-5%20%28File%20responses%29/Slides%20in%20PDF%20%28Please%20name%20the%20submitted%20file%20as%20_firstname lastname slides.pdf%29.%20%28File%20responses%29/27-Michael%20Jordan-Session%20V.pdf#page=15
[00:28:40] Michael I. Jordan Session V Slides
https://rdi.berkeley.edu/events/sbc-assets/pdfs/Summit%20session%20speaker%20slides%20submission%20form-s1-5%20%28File%20responses%29/Slides%20in%20PDF%20%28Please%20name%20the%20submitted%20file%20as%20_firstname lastname slides.pdf%29.%20%28File%20responses%29/27-Michael%20Jordan-Session%20V.pdf#page=1
[01:08:13] Three Foundational Disciplines
https://rdi.berkeley.edu/events/sbc-assets/pdfs/Summit%20session%20speaker%20slides%20submission%20form-s1-5%20%28File%20responses%29/Slides%20in%20PDF%20%28Please%20name%20the%20submitted%20file%20as%20_firstname lastname slides.pdf%29.%20%28File%20responses%29/27-Michael%20Jordan-Session%20V.pdf#page=35
organization:
[00:45:58] UnitedMasters
https://www.unitedmasters.com/
book:
[00:48:30] Human Compatible: Artificial Intelligence and the Problem of Control
https://aima.cs.berkeley.edu/~russell/hc.html
[01:00:56] Theory of Games and Economic Behavior
https://press.princeton.edu/books/paperback/9780691130613/theory-of-games-and-economic-behavior Intelligence is collective, not artificial — Prof. Michael I. Jordan (UC Berkeley / Inria)](https://i.ytimg.com/vi/AREWYbVtX64/mqdefault.jpg)
![The Final Boss of Deep Learning
We often think of Large Language Models (LLMs) as all-knowing, but as the team reveals, they still struggle with the logic of a second-grader. Why can’t ChatGPT reliably add large numbers? Why does it hallucinate the laws of physics? The answer lies in the architecture. This episode explores how *Category Theory*—an ultra-abstract branch of mathematics—could provide the Periodic Table for neural networks, turning the alchemy of modern AI into a rigorous science.
In this deep-dive exploration, *Andrew Dudzik*, *Petar Velichkovich*, *Taco Cohen*, *Bruno Gavranović*, and *Paul Lessard* join host *Tim Scarfe* to discuss the fundamental limitations of today’s AI and the radical mathematical framework that might fix them.
Key Insights in This Episode:
* *The Addition Problem:* *Andrew Dudzik* explains why LLMs dont actually know math—they just recognize patterns. When you change a single digit in a long string of numbers, the pattern breaks because the model lacks the internal machinery to perform a simple carry operation.
* *Beyond Alchemy:* *Tim Scarfe* argues that deep learning is currently in its alchemy phase—we have powerful results, but we lack a unifying theory. Category Theory is proposed as the framework to move AI from trial-and-error to principled engineering. [00:13:49]
* *Algebra with Colors:* To make Category Theory accessible, the guests use brilliant analogies—like thinking of matrices as *magnets with colors* that only snap together when the types match. This partial compositionality is the secret to building more complex internal reasoning. [00:09:17]
* *Synthetic vs. Analytic Math:* *Paul Lessard* breaks down the philosophical shift needed in AI research: moving from Analytic math (what things are made of) to Synthetic math (how things behave and relate to one another). [00:23:41]
* *The 4D Carry:* In a mind-blowing conclusion, the team discusses how simple algorithmic tasks, like carrying the one in addition, actually relate to complex geometric structures like *Hopf Fibrations*. [00:39:30]
Why This Matters for AGI
If we want AI to solve the worlds hardest scientific problems, it cant just be a stochastic parrot. It needs to internalize the rules of logic and computation. By imbuing neural networks with categorical priors, researchers are attempting to build a future where AI doesnt just predict the next word—it understands the underlying structure of the universe.
TIMESTAMPS:
00:00:00 The Failure of LLM Addition & Physics
00:01:26 Tool Use vs Intrinsic Model Quality
00:03:07 Efficiency Gains via Internalization
00:04:28 Geometric Deep Learning & Equivariance
00:07:05 Limitations of Group Theory
00:09:17 Category Theory: Algebra with Colors
00:11:25 The Systematic Guide of Lego-like Math
00:13:49 The Alchemy Analogy & Unifying Theory
00:15:33 Information Destruction & Reasoning
00:18:00 Pathfinding & Monoids in Computation
00:20:15 System 2 Reasoning & Error Awareness
00:23:31 Analytic vs Synthetic Mathematics
00:25:52 Morphisms & Weight Tying Basics
00:26:48 2-Categories & Weight Sharing Theory
00:28:55 Higher Categories & Emergence
00:31:41 Compositionality & Recursive Folds
00:34:05 Syntax vs Semantics in Network Design
00:36:14 Homomorphisms & Multi-Sorted Syntax
00:39:30 The Carrying Problem & Hopf Fibrations
REFERENCES:
Company:
Model:
[00:01:05] Veo
https://deepmind.google/models/veo/
[00:01:10] Genie
https://deepmind.google/blog/genie-3-a-new-frontier-for-world-models/
Paper:
[00:04:30] Geometric Deep Learning Blueprint
https://arxiv.org/abs/2104.13478
[00:16:45] AlphaGeometry
https://deepmind.google/blog/alphageometry-an-olympiad-level-ai-system-for-geometry/
[00:16:55] AlphaCode
https://arxiv.org/abs/2203.07814
[00:17:05] FunSearch
https://www.nature.com/articles/s41586-023-06924-6
[00:37:00] Attention Is All You Need
https://arxiv.org/abs/1706.03762
[00:43:00] Categorical Deep Learning
https://arxiv.org/abs/2402.15332
Thanks to ly3xqhl8g9 on our Discord server for the draft show review! The Final Boss of Deep Learning](https://i.ytimg.com/vi/AWqvBdqCAAE/mqdefault.jpg)

![AI in 2024 - efficiency over model size (Nick Jakobi)
This video is sponsored by Cohere
Nick Jakobi, Director of Product for the modeling team at Cohere, discusses the practical realities of deploying large language models in enterprise environments. The conversation covers how Cohere differentiates in a crowded LLM market through enterprise-focused capabilities like retrieval-augmented generation (RAG), tool use, and multilingual support, rather than competing purely on model scale.
Jakobi offers a candid perspective on the economics of model selection: most enterprise customers gravitate toward the smallest model that solves their use case, making cost and latency as important as raw capability. He describes the emerging token game where model providers compete on price-performance ratios, and argues that making medium-sized models cheaper and faster may be more commercially valuable than pushing the frontier of model intelligence.
The discussion extends into broader questions about how generative AI is reshaping education, workforce dynamics, and information ecosystems. Jakobi shares personal reflections on whether traditional education remains relevant when AI tools are becoming ubiquitous, and offers a measured take on AI safety that centers on the practical challenges of disinformation and autonomous agency rather than speculative existential risks.
REFERENCES:
Company:
[00:00:00] Cohere
https://cohere.com/
Paper:
[00:00:00] Episode Shownotes
https://www.dropbox.com/scl/fi/sjwqllqh1ughuu8ismonk/NickJakobi2.pdf?rlkey=d20wydhhala2hmh4yaoo55wyy&st=unso2w6i&dl=0
[00:00:30] RAG Paper
https://arxiv.org/abs/2005.11401
Person:
[00:00:20] Nick Jakobi
https://www.linkedin.com/in/nickjakobi/
Product:
[00:14:50] Cohere Command R
https://cohere.com/command
Tool:
[00:20:50] LMSYS Chatbot Arena
https://chat.lmsys.org/
LINKS:
Full Transcript: https://app.rescript.info/share/d694e87868aa6d0272fe26e0a7b4de87
Download PDF transcript: https://app.rescript.info/api/public/sessions/72c99a6b30e9b8d0/pdf AI in 2024 - efficiency over model size (Nick Jakobi)](https://i.ytimg.com/vi/BJMi_dAwE8k/mqdefault.jpg)


![Test-Time Adaptation: A New Frontier in AI [Dr. Jonas Hübotter]
Jonas Hübotter, PhD student at ETH Zurichs Institute for Machine Learning, discusses his groundbreaking research on test-time computation and local learning. He demonstrates how smaller models can outperform larger ones by 30x through strategic test-time computation and introduces a novel paradigm combining inductive and transductive learning approaches.
Using Bayesian linear regression as a surrogate model for uncertainty estimation, Jonas explains how models can efficiently adapt to specific tasks without massive pre-training. He draws an analogy to Google Earths variable resolution system to illustrate dynamic resource allocation based on task complexity.
The conversation explores the future of AI architecture, envisioning systems that continuously learn and adapt beyond current monolithic models. Jonas concludes by proposing hybrid deployment strategies combining local and cloud computation, suggesting a future where compute resources are allocated based on task complexity rather than fixed model size.
This research represents a significant shift in machine learning, prioritizing intelligent resource allocation and adaptive learning over traditional scaling approaches.
SPONSOR MESSAGES:
CentML offers competitive pricing for GenAI model deployment, with flexible options to suit a wide range of models, from small to large-scale deployments.
https://centml.ai/pricing/
Tufa AI Labs is a brand new research lab in Zurich started by Benjamin Crouzier focussed on ARC and AGI, they just acquired MindsAI - the current winners of the ARC challenge. Are you interested in working on ARC, or getting involved in their events? Goto https://tufalabs.ai/
Transcription, references and show notes PDF download:
https://www.dropbox.com/scl/fi/cxg80p388snwt6qbp4m52/JonasFinal.pdf?rlkey=glk9mhpzjvesanlc14rtpvk4r&st=6qwi8n3x&dl=0
Jonas Hübotter
https://jonhue.github.io/
https://scholar.google.com/citations?user=pxi_RkwAAAAJ
Transductive Active Learning: Theory and Applications (NeurIPS 2024)
https://arxiv.org/pdf/2402.15898
EFFICIENTLY LEARNING AT TEST-TIME: ACTIVE FINE-TUNING OF LLMS (SIFT)
https://arxiv.org/pdf/2410.08020
TOC:
1. Test-Time Computation Fundamentals
[00:00:00] Intro
[00:03:10] 1.1 Test-Time Computation and Model Performance Comparison
[00:05:52] 1.2 Retrieval Augmentation and Machine Teaching Strategies
[00:09:40] 1.3 In-Context Learning vs Fine-Tuning Trade-offs
2. System Architecture and Intelligence
[00:15:58] 2.1 System Architecture and Intelligence Emergence
[00:23:22] 2.2 Active Inference and Constrained Agency in AI
[00:29:52] 2.3 Evolution of Local Learning Methods
[00:32:05] 2.4 Vapniks Contributions to Transductive Learning
3. Resource Optimization and Local Learning
[00:34:35] 3.1 Computational Resource Allocation in ML Models
[00:35:30] 3.2 Historical Context and Traditional ML Optimization
[00:37:55] 3.3 Variable Resolution Processing and Active Inference in ML
[00:43:01] 3.4 Local Learning and Base Model Capacity Trade-offs
[00:48:04] 3.5 Active Learning vs Local Learning Approaches
4. Information Retrieval and Model Interpretability
[00:51:08] 4.1 Information Retrieval and Nearest Neighbor Limitations
[01:03:07] 4.2 Model Interpretability and Surrogate Models
[01:15:03] 4.3 Bayesian Uncertainty Estimation and Surrogate Models
5. Distributed Systems and Deployment
[01:23:56] 5.1 Memory Architecture and Controller Systems
[01:28:14] 5.2 Evolution from Static to Distributed Learning Systems
[01:38:03] 5.3 Transductive Learning and Model Specialization
[01:41:58] 5.4 Hybrid Local-Cloud Deployment Strategies Test-Time Adaptation: A New Frontier in AI [Dr. Jonas Hübotter]](https://i.ytimg.com/vi/C6sSs6NgANo/mqdefault.jpg)

![The Gap Between Humans and Machines Is [Dr. Max Bartolo]
SPONSOR MESSAGES:
***
Tufa AI Labs is a brand new research lab in Zurich started by Benjamin Crouzier focussed on o-series style reasoning and AGI. They are hiring a Chief Engineer and ML engineers. Events in Zurich.
Dr. Max Bartolo from Cohere discusses the gap between model capabilities and genuine robustness: why next-token prediction can produce impressive results yet still fail on slightly reformulated…
TIMESTAMPS:
00:00:00 Model Reasoning and Consistency Verification
00:03:25 Influence Functions and Distributed Knowledge Analysis
00:10:28 AI Application Development and Model Deployment
00:14:24 AI Alignment and Human Feedback Limitations
00:20:15 Human Evaluation Challenges and Factuality Assessment
00:27:15 Cultural and Demographic Influences on Model Behavior
00:32:43 Adversarial Examples and Model Robustness
00:41:54 DynaBench and Dynamic Benchmarking Approaches
00:50:02 Benchmarking Challenges and Data-Centric Evaluation
00:55:15 Cohere Command A Development Process
01:00:26 Model Quantization and Performance Evaluation
01:05:18 Reasoning Capabilities and Training Progression
01:13:48 Context Windows and Enterprise Applications
REFERENCES:
person:
[00:00:00] Max Bartolo Website
https://www.maxbartolo.com/
company:
[00:00:00] Cohere
https://cohere.com/command
[00:12:10] Command A Model
https://huggingface.co/CohereForAI/c4ai-command-a-03-2025
paper:
[00:03:25] Procedural Knowledge in Pretraining Drives Reasoning in LLMs
https://cohere.com/research/papers/procedural-knowledge-in-pretraining-drives-reasoning-in-large-language-models-2024-11-20
[00:04:15] Influence Functions in Machine Learning
https://arxiv.org/abs/1703.04730
[00:08:05] Studying Large Language Model Generalization with Influence Functions
https://arxiv.org/abs/2308.03296
[00:16:15] Human Feedback is not Gold Standard
https://arxiv.org/abs/2309.16349
[00:27:15] The PRISM Alignment Dataset
https://arxiv.org/abs/2404.16019
[00:32:50] Adversarial Examples Are Not Bugs, They Are Features
https://arxiv.org/abs/1905.02175
[00:43:00] DynaBench: Rethinking Benchmarking in NLP
https://aclanthology.org/2021.naacl-main.324.pdf
[00:50:15] Sara Hooker on Compute Limitations
https://arxiv.org/html/2407.05694v1
[00:53:25] DataPerf: Benchmarks for Data-Centric AI
https://arxiv.org/abs/2207.10062
[01:04:35] DROP: A Reading Comprehension Benchmark
https://arxiv.org/abs/1903.00161
[01:07:05] GSM8k
https://paperswithcode.com/sota/arithmetic-reasoning-on-gsm8k
[01:09:30] ARC-AGI Challenge
https://github.com/fchollet/ARC-AGI
LINKS:
Full Transcript: https://app.rescript.info/share/163bf5e7338685f635fc0b8d6920005c
Download PDF transcript: https://app.rescript.info/api/public/sessions/7167a679366f97f8/pdf
REFS:
[00:03:10] Research at Cohere with Laura Ruis et al., Max Bartolo, Laura Ruis et al.
https://cohere.com/research/papers/procedural-knowledge-in-pretraining-drives-reasoning-in-large-language-models-2024-11-20
[00:04:15] Influence functions in machine learning, Koh & Liang
https://arxiv.org/abs/1703.04730
[00:08:05] Studying Large Language Model Generalization with Influence Functions, Roger Grosse et al.
https://storage.prod.researchhub.com/uploads/papers/2023/08/08/2308.03296.pdf
[00:11:10] The LLM ARChitect: Solving ARC-AGI Is A Matter of Perspective, Daniel Franzen, Jan Disselhoff, and David Hartmann
https://github.com/da-fr/arc-prize-2024/blob/main/the_architects.pdf
[00:12:10] Hugging Face model repo for C4AI Command A, Cohere and Cohere For AI
https://huggingface.co/CohereForAI/c4ai-command-a-03-2025
[00:13:30] OpenInterpreter
https://github.com/KillianLucas/open-interpreter
[00:16:15] Human Feedback is not Gold Standard, Tom Hosking, Max Bartolo, Phil Blunsom
https://arxiv.org/abs/2309.16349
[00:27:15] The PRISM Alignment Dataset, Hannah Kirk et al.
https://arxiv.org/abs/2404.16019
[00:32:50] How adversarial examples arise, Andrew Ilyas, Shibani Santurkar, Dimitris Tsipras, Logan Engstrom, Brandon Tran, Aleksander Madry
https://arxiv.org/abs/1905.02175
[00:43:00] DynaBench platform paper, Douwe Kiela et al.
https://aclanthology.org/2021.naacl-main.324.pdf
[00:50:15] Sara Hookers work on compute limitations, Sara Hooker
https://arxiv.org/html/2407.05694v1
[00:53:25] DataPerf: Community-led benchmark suite, Mazumder et al.
https://arxiv.org/abs/2207.10062
[01:04:35] DROP, Dheeru Dua et al.
https://arxiv.org/abs/1903.00161
[01:07:05] GSM8k, Cobbe et al.
https://paperswithcode.com/sota/arithmetic-reasoning-on-gsm8k
[01:09:30] ARC, François Chollet
https://github.com/fchollet/ARC-AGI
[01:15:50] Command A, Cohere
https://cohere.com/blog/command-a
[01:22:55] Enterprise search using LLMs, Cohere
https://cohere.com/blog/commonly-asked-questions-about-search-from-coheres-enterprise-customers The Gap Between Humans and Machines Is [Dr. Max Bartolo]](https://i.ytimg.com/vi/DL7qwmWWk88/mqdefault.jpg)