Sign up to save your podcasts
Or
Share AI Post Transformers
Share to email
Share to Facebook
Share to X
Share to email
Share to Facebook
Share to X
Or
AI Post Transformers
By mcgrofTechnology
AI Post Transformers episodes
- Jev-Mem: Fast System-One Controllers for Agentic MemoryThis episode explores Jev-Mem, a design for agent memory that moves small management decisions, such as classifying a memory, linking it to others, and deciding when to stop searching, out of the language model's text generation and into a fast controller that outputs labels and scores. It walks through why agents need external memory despite long context windows (RULER's finding of shorter effective context, and Lost in the Middle). It traces the progression from plain retrieval through MemGPT-style paging to multi-graph memory, using a running example of a bicycle purchase that needs temporal, causal, and entity links to answer correctly. The discussion frames the "System One / System Two" split as an analogy for allocating compute rather than a claim about brains. It separates the approach from distilling reasoning into a single model, and sets up a three-layer architecture of control, memory, and reasoning planes. It also flags what to scrutinize: how much of the gain comes from the graph versus the controller, and that a closely related system is cited but not benchmarked. Listeners interested in cutting the token cost of agent memory will find the long-conversation benchmark setup and the typed yes/no query interface of the controller especially concrete.Sources:1. Jev-Mem: System-One-Controlled Agentic Memory for Efficient AI Agents — Dongming Jiang, Yi Li, Bingzhe Li, 2026http://arxiv.org/abs/2609.239862. Dual-Processing Accounts of Reasoning, Judgment, and Social Cognition — Jonathan St. B. T. Evans, 2008https://scholar.google.com/scholar?q=Dual-Processing+Accounts+of+Reasoning%2C+Judgment%2C+and+Social+Cognition3. Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory — Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, Deshraj Yadav, 2025https://scholar.google.com/scholar?q=Mem0%3A+Building+Production-Ready+AI+Agents+with+Scalable+Long-Term+Memory4. MemGPT: Towards LLMs as Operating Systems — Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G. Patil, Ion Stoica, Joseph E. Gonzalez, 2023https://scholar.google.com/scholar?q=MemGPT%3A+Towards+LLMs+as+Operating+Systems5. Adaptive-RAG: Learning to Adapt Retrieval-Augmented Large Language Models through Question Complexity — Soyeong Jeong, Jinheon Baek, Sukmin Cho, Sung Ju Hwang, Jong C. Park, 2024https://scholar.google.com/scholar?q=Adaptive-RAG%3A+Learning+to+Adapt+Retrieval-Augmented+Large+Language+Models+through+Question+Complexity6. Distilling System 2 into System 1 — Ping Yu, Jing Xu, Jason Weston, Ilia Kulikov, 2024https://scholar.google.com/scholar?q=Distilling+System+2+into+System+17. Zero-Mem: Zero-token memory operations for LLM agents — Yilin Xiao et al., 2026https://scholar.google.com/scholar?q=Zero-Mem%3A+Zero-token+memory+operations+for+LLM+agents8. MAGMA: A Multi-Graph based Agentic Memory Architecture for AI Agents — Dongming Jiang, Yi Li, Guanpeng Li, Bingzhe Li, 2026https://scholar.google.com/scholar?q=MAGMA%3A+A+Multi-Graph+based+Agentic+Memory+Architecture+for+AI+Agents9. HAGE: Harnessing Agentic Memory via RL-driven Weighted Graph Evolution — Dongming Jiang et al., 2026https://scholar.google.com/scholar?q=HAGE%3A+Harnessing+Agentic+Memory+via+RL-driven+Weighted+Graph+Evolution10. Self-RAG: Learning to Retrieve, Generate, and Critique through Self-Reflection — Akari Asai, Zeqiu Wu, Yizhong Wang, Avirup Sil, Hannaneh Hajishirzi, 2023https://scholar.google.com/scholar?q=Self-RAG%3A+Learning+to+Retrieve%2C+Generate%2C+and+Critique+through+Self-Reflection11. Adaptive-RAG: Learning to Adapt Retrieval-Augmented LLMs through Question Complexity — Soyeong Jeong, Jinheon Baek, Sukmin Cho, Sung Ju Hwang, Jong C. Park, 2024https://scholar.google.com/scholar?q=Adaptive-RAG%3A+Learning+to+Adapt+Retrieval-Augmented+LLMs+through+Question+Complexity12. FrugalGPT and RouteLLM — Lingjiao Chen, Matei Zaharia, James Zou (FrugalGPT); Isaac Ong et al. (RouteLLM), 2024 / 2025https://scholar.google.com/scholar?q=FrugalGPT+and+RouteLLM13. Evaluating Very Long-Term Conversational Memory of LLM Agents (LoCoMo) — Adyasha Maharana et al., 2024https://scholar.google.com/scholar?q=Evaluating+Very+Long-Term+Conversational+Memory+of+LLM+Agents+%28LoCoMo%2914. LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory — Di Wu et al., 2024https://scholar.google.com/scholar?q=LongMemEval%3A+Benchmarking+Chat+Assistants+on+Long-Term+Interactive+Memory15. HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models — Bernal Jimenez Gutierrez et al., 2024https://scholar.google.com/scholar?q=HippoRAG%3A+Neurobiologically+Inspired+Long-Term+Memory+for+Large+Language+ModelsInteractive Visualization: Jev-Mem: Fast System-One Controllers for Agentic Memory17 min
- Cogentic: Multi-Agent Harness Discovers Proofs for Open ProblemsThis episode examines "Cogentic," a multi-agent harness that reportedly produced new, expert-verified proofs for five open problems in online learning, auction theory, and mechanism design, using about a hundred model calls on most problems and about a thousand on the hardest. It places the work among three approaches: formal theorem provers like Lean, search-for-objects systems like FunSearch and AlphaEvolve that need a computable score, and inference-time scaling, where coverage only becomes an answer if something reliably picks the right candidate. The discussion walks through the harness design. An orchestrator assigns directions, parallel prover agents draft proofs, per-prover summarizers give each a different reading of the history, and adversarial per-draft and cross-draft verifiers must both pass a result. It also raises open questions: the verifiers are prompted copies of the same model family as the provers, nothing is machine-checked, the models are proprietary, and the number of failed attempts behind the five wins and the cost are unclear. A listener would find it useful for judging whether the harness itself, or just a strong model with patience, deserves the credit.Sources:1. Cogentic: Multi-Agent Orchestration for Automated Proof Discovery — Yang Cai, Vineet Gupta, Yanchen Jiang, Christopher Liaw, Aranyak Mehta, Grigoris Velegkas, Di Wang, 2026http://arxiv.org/abs/2609.403242. Mathematical discoveries from program search with large language models (FunSearch) — Bernardino Romera-Paredes, Mohammadamin Barekatain, Alexander Novikov, et al., 2024https://scholar.google.com/scholar?q=Mathematical+discoveries+from+program+search+with+large+language+models+%28FunSearch%293. AlphaEvolve: A coding agent for scientific and algorithmic discovery — Alexander Novikov, Ngân Vũ, Marvin Eisenberger, et al., 2025https://scholar.google.com/scholar?q=AlphaEvolve%3A+A+coding+agent+for+scientific+and+algorithmic+discovery4. LeanDojo: Theorem Proving with Retrieval-Augmented Language Models — Kaiyu Yang, Aidan Swope, Alex Gu, Rahul Chalamala, et al., 2023https://scholar.google.com/scholar?q=LeanDojo%3A+Theorem+Proving+with+Retrieval-Augmented+Language+Models5. Draft, Sketch, and Prove: Guiding Formal Theorem Provers with Informal Proofs — Albert Q. Jiang, Sean Welleck, Jin Peng Zhou, et al., 2022https://scholar.google.com/scholar?q=Draft%2C+Sketch%2C+and+Prove%3A+Guiding+Formal+Theorem+Provers+with+Informal+Proofs6. Large Language Monkeys: Scaling Inference Compute with Repeated Sampling — Bradley Brown, Jordan Juravsky, Ryan Ehrlich, et al., 2024https://scholar.google.com/scholar?q=Large+Language+Monkeys%3A+Scaling+Inference+Compute+with+Repeated+Sampling7. Scaling LLM Test-Time Compute Optimally can be More Effective than Scaling Model Parameters — Charlie Snell, Jaehoon Lee, Kelvin Xu, Aviral Kumar, 2024https://scholar.google.com/scholar?q=Scaling+LLM+Test-Time+Compute+Optimally+can+be+More+Effective+than+Scaling+Model+Parameters8. Improving Factuality and Reasoning in Language Models through Multiagent Debate — Yilun Du, Shuang Li, Antonio Torralba, Joshua B. Tenenbaum, Igor Mordatch, 2024https://scholar.google.com/scholar?q=Improving+Factuality+and+Reasoning+in+Language+Models+through+Multiagent+Debate9. Mathematical discoveries from program search with large language models — Romera-Paredes et al., 2024https://scholar.google.com/scholar?q=Mathematical+discoveries+from+program+search+with+large+language+models10. Let's Verify Step by Step — Hunter Lightman, Vineet Kosaraju, Yura Burda, et al., 2023https://scholar.google.com/scholar?q=Let%27s+Verify+Step+by+Step11. Training Verifiers to Solve Math Word Problems — Karl Cobbe, Vineet Kosaraju, Mohammad Bavarian, et al., 2021https://scholar.google.com/scholar?q=Training+Verifiers+to+Solve+Math+Word+Problems12. Large Language Models Cannot Self-Correct Reasoning Yet — Jie Huang, Xinyun Chen, Swaroop Mishra, et al., 2024https://scholar.google.com/scholar?q=Large+Language+Models+Cannot+Self-Correct+Reasoning+Yet13. AlphaProof / Olympiad-level formal mathematical reasoning with reinforcement learning — Thomas Hubert et al. (Google DeepMind), 2026https://scholar.google.com/scholar?q=AlphaProof+%2F+Olympiad-level+formal+mathematical+reasoning+with+reinforcement+learning14. Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena — Zheng et al., 2023https://scholar.google.com/scholar?q=Judging+LLM-as-a-Judge+with+MT-Bench+and+Chatbot+Arena15. Mathematical exploration and discovery at scale — Georgiev, Gómez-Serrano, Tao, Wagner, 2025https://scholar.google.com/scholar?q=Mathematical+exploration+and+discovery+at+scale16. Stellar Colosseum; ProofCouncil; AI co-mathematician; Towards autonomous mathematics research — Lin et al.; Schmitt et al.; Zheng et al.; Feng et al., 2026https://scholar.google.com/scholar?q=Stellar+Colosseum%3B+ProofCouncil%3B+AI+co-mathematician%3B+Towards+autonomous+mathematics+research17. Multiscale reward hedging from correct demonstrations / Tight regret bound for online inverse linear optimization via multiscale matrix weights — Dewasurendra; Sakaue, 2026https://scholar.google.com/scholar?q=Multiscale+reward+hedging+from+correct+demonstrations+%2F+Tight+regret+bound+for+online+inverse+linear+optimization+via+multiscale+matrix+weights18. Agent-as-a-Judge: Evaluate Agents with Agents — Zhuge et al., 2024https://scholar.google.com/scholar?q=Agent-as-a-Judge%3A+Evaluate+Agents+with+AgentsInteractive Visualization: Cogentic: Multi-Agent Harness Discovers Proofs for Open Problems17 min
- Code Predicts, Jev Judges: Separating Simulation from EvaluationThis episode examines a paper on a closed, API-only action-selection model (Jev) that returns a probability for each option in a single call, with no reasoning text. The paper's question is which decisions such a model can be trusted with. It separates "evaluation" decisions, where the answer can be read from the input and options, from "simulation" decisions, where the right option depends on an unstated prediction such as an opponent's move or a required subgoal like washing a knife before placing it. The discussion works through the paper's test setups, random three-by-three payoff games with iterated dominance depth and the ALFWorld household text game. It also covers the "lure" idea: a fixed surface shortcut that sometimes points to the wrong option, which separates real competence from luck. The paper's proposed fix is to have code do the predicting and the model do the judging. Listeners get a clear framework for why a model can ace trap questions yet fail a simple household step, and for what that means for building cheap, fast agent loops.Sources:1. Code Owns the Simulation, Jev Owns the Evaluation — Yaodong Yang, Hongyao Tang, Yi Ma, Xingyu Fan, Weixun Wang, Jinpeng Li, Tianpei Yang, 2026http://arxiv.org/abs/2610.018342. Thinking, Fast and Slow (book) / Dual-Process Theories of Higher Cognition: Advancing the Debate — Daniel Kahneman (2011); Jonathan St. B. T. Evans and Keith E. Stanovich (2013), 2011 / 2013https://scholar.google.com/scholar?q=Thinking%2C+Fast+and+Slow+%28book%29+%2F+Dual-Process+Theories+of+Higher+Cognition%3A+Advancing+the+Debate3. Cognitive Reflection and Decision Making (the Cognitive Reflection Test); Human-like intuitive behavior and reasoning biases emerged in large language models but disappeared in ChatGPT — Shane Frederick (2005); Thilo Hagendorff, Sarah Fabi, Michal Kosinski (2023), 2005 / 2023https://scholar.google.com/scholar?q=Cognitive+Reflection+and+Decision+Making+%28the+Cognitive+Reflection+Test%29%3B+Human-like+intuitive+behavior+and+reasoning+biases+emerged+in+large+language+models+but+disappeared+in+ChatGPT4. Do As I Can, Not As I Say: Grounding Language in Robotic Affordances (SayCan) — Michael Ahn, Anthony Brohan, et al. (Google Robotics / Everyday Robots), 2022https://scholar.google.com/scholar?q=Do+As+I+Can%2C+Not+As+I+Say%3A+Grounding+Language+in+Robotic+Affordances+%28SayCan%295. SwiftSage: A Generative Agent with Fast and Slow Thinking for Complex Interactive Tasks — Bill Yuchen Lin, Yicheng Fu, Karina Yang, et al., 2023https://scholar.google.com/scholar?q=SwiftSage%3A+A+Generative+Agent+with+Fast+and+Slow+Thinking+for+Complex+Interactive+Tasks6. Measuring and Narrowing the Compositionality Gap in Language Models — Ofir Press, Muru Zhang, Sewon Min, Ludwig Schmidt, Noah A. Smith, Mike Lewis, 2022https://scholar.google.com/scholar?q=Measuring+and+Narrowing+the+Compositionality+Gap+in+Language+Models7. Can Large Language Models Serve as Rational Players in Game Theory? A Systematic Analysis; Playing Repeated Games with Large Language Models — Caoyun Fan, Jindou Chen, Yaohui Jin, Hao He (2023); Elif Akata, Lion Schulz, et al. (2023), 2023https://scholar.google.com/scholar?q=Can+Large+Language+Models+Serve+as+Rational+Players+in+Game+Theory%3F+A+Systematic+Analysis%3B+Playing+Repeated+Games+with+Large+Language+Models8. LLMs Can't Plan, But Can Help Planning in LLM-Modulo Frameworks (and On the Planning Abilities of Large Language Models) — Subbarao Kambhampati et al. (2024); Karthik Valmeekam et al. (2023), 2023-2024https://scholar.google.com/scholar?q=LLMs+Can%27t+Plan%2C+But+Can+Help+Planning+in+LLM-Modulo+Frameworks+%28and+On+the+Planning+Abilities+of+Large+Language+Models%299. Mind's Eye: Grounded Language Model Reasoning through Simulation — Ruibo Liu, Jason Wei, Shixiang Shane Gu, et al., 2022https://scholar.google.com/scholar?q=Mind%27s+Eye%3A+Grounded+Language+Model+Reasoning+through+Simulation10. Reasoning with Language Model is Planning with World Model (RAP) — Shibo Hao, Yi Gu, Haodi Ma, et al., 2023https://scholar.google.com/scholar?q=Reasoning+with+Language+Model+is+Planning+with+World+Model+%28RAP%2911. Code as Policies: Language Model Programs for Embodied Control; Language to Rewards for Robotic Skill Synthesis — Jacky Liang, Wenlong Huang, et al. (2022); Wenhao Yu, Nimrod Gileadi, et al. (2023), 2022-2023https://scholar.google.com/scholar?q=Code+as+Policies%3A+Language+Model+Programs+for+Embodied+Control%3B+Language+to+Rewards+for+Robotic+Skill+Synthesis12. Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model (MuZero) — Julian Schrittwieser, Ioannis Antonoglou, Thomas Hubert, David Silver, et al., 2020https://scholar.google.com/scholar?q=Mastering+Atari%2C+Go%2C+Chess+and+Shogi+by+Planning+with+a+Learned+Model+%28MuZero%2913. ALFWorld: Aligning Text and Embodied Environments for Interactive Learning — Mohit Shridhar, Xingdi Yuan, Marc-Alexandre Côté, Yonatan Bisk, Adam Trischler, Matthew Hausknecht, 2021https://scholar.google.com/scholar?q=ALFWorld%3A+Aligning+Text+and+Embodied+Environments+for+Interactive+Learning14. ReAct: Synergizing Reasoning and Acting in Language Models — Shunyu Yao, Jeffrey Zhao, Dian Yu, Nan Du, Izhak Shafran, Karthik Narasimhan, Yuan Cao, 2022https://scholar.google.com/scholar?q=ReAct%3A+Synergizing+Reasoning+and+Acting+in+Language+Models15. Reflexion: Language Agents with Verbal Reinforcement Learning — Noah Shinn, Federico Cassano, Ashwin Gopinath, Karthik Narasimhan, Shunyu Yao, 2023https://scholar.google.com/scholar?q=Reflexion%3A+Language+Agents+with+Verbal+Reinforcement+Learning16. Language Agent Tree Search Unifies Reasoning, Acting, and Planning in Language Models (LATS) — Andy Zhou, Kai Yan, Michal Shlapentokh-Rothman, Haohan Wang, Yu-Xiong Wang, 2023https://scholar.google.com/scholar?q=Language+Agent+Tree+Search+Unifies+Reasoning%2C+Acting%2C+and+Planning+in+Language+Models+%28LATS%2917. Faith and Fate: Limits of Transformers on Compositionality — Dziri et al., 2023https://scholar.google.com/scholar?q=Faith+and+Fate%3A+Limits+of+Transformers+on+Compositionality18. Do Large Language Models Latently Perform Multi-Hop Reasoning? — Yang et al., 2024https://scholar.google.com/scholar?q=Do+Large+Language+Models+Latently+Perform+Multi-Hop+Reasoning%3F19. The Expressive Power of Transformers with Chain of Thought / Chain of Thought Empowers Transformers to Solve Inherently Serial Problems — Merrill & Sabharwal; Li et al., 2023-2024https://scholar.google.com/scholar?q=The+Expressive+Power+of+Transformers+with+Chain+of+Thought+%2F+Chain+of+Thought+Empowers+Transformers+to+Solve+Inherently+Serial+Problems20. Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model (MuZero) / Mastering the game of Go without human knowledge (AlphaZero) — Schrittwieser et al.; Silver et al., 2020 / 2017https://scholar.google.com/scholar?q=Mastering+Atari%2C+Go%2C+Chess+and+Shogi+by+Planning+with+a+Learned+Model+%28MuZero%29+%2F+Mastering+the+game+of+Go+without+human+knowledge+%28AlphaZero%2921. Large Language Models Still Can't Plan / On the Planning Abilities of Large Language Models (PlanBench) — Valmeekam et al., 2023https://scholar.google.com/scholar?q=Large+Language+Models+Still+Can%27t+Plan+%2F+On+the+Planning+Abilities+of+Large+Language+Models+%28PlanBench%2922. Language Models Don't Always Say What They Think / Language Models (Mostly) Know What They Know — Turpin et al.; Kadavath et al., 2023 / 2022https://scholar.google.com/scholar?q=Language+Models+Don%27t+Always+Say+What+They+Think+%2F+Language+Models+%28Mostly%29+Know+What+They+Know23. Generative Verifiers: Reward Modeling as Next-Token Prediction — Zhang et al., 2024https://scholar.google.com/scholar?q=Generative+Verifiers%3A+Reward+Modeling+as+Next-Token+Prediction24. Large Language Models Are Not Robust Multiple Choice Selectors — Zheng et al., 2024https://scholar.google.com/scholar?q=Large+Language+Models+Are+Not+Robust+Multiple+Choice+Selectors25. Dreamer / DreamerV3 — Hafner et al., 2020-2023https://scholar.google.com/scholar?q=Dreamer+%2F+DreamerV3Interactive Visualization: Code Predicts, Jev Judges: Separating Simulation from Evaluation20 min
- Measuring Emergent Coordination in Multi-Agent LLMs via Information DecompositionThis episode examines a 2026 ICLR paper that asks whether groups of LLM agents form genuine collectives or are just aggregates of individuals. Ten GPT-4.1 agents play a group binary search game, each privately guessing a number so the total hits a hidden target while seeing only "too high" or "too low" feedback. The only way to succeed is to take complementary offsets, since identical strategies make the group overshoot and undershoot in a loop. The hosts explain partial information decomposition, which splits information into unique, redundant, and synergistic parts. They note that the results depend on which redundancy definition is chosen, and that the leading alternative measure isn't reported. They also walk through the three prompt conditions (plain, persona, and persona plus a one-sentence theory-of-mind nudge), the emergence and coalition tests applied to short trajectories, and the statistical tension between Fisher's method and Wilcoxon tests. Listeners get a skeptical look at how much evidence supports the claim that prompting can turn a group of agents into a higher-order collective.Sources:1. Emergent Coordination in Multi-Agent Language Models — Christoph Riedl, 2025http://arxiv.org/abs/2510.051742. Nonnegative Decomposition of Multivariate Information — Paul L. Williams, Randall D. Beer, 2010https://scholar.google.com/scholar?q=Nonnegative+Decomposition+of+Multivariate+Information3. Reconciling emergences: An information-theoretic approach to identify causal emergence in multivariate data — Fernando E. Rosas, Pedro A. M. Mediano, Henrik J. Jensen, Anil K. Seth, Adam B. Barrett, Robin L. Carhart-Harris, Daniel Bor, 2020https://scholar.google.com/scholar?q=Reconciling+emergences%3A+An+information-theoretic+approach+to+identify+causal+emergence+in+multivariate+data4. Greater than the parts: a review of the information decomposition approach to causal emergence — Pedro A. M. Mediano, Fernando E. Rosas, Andrea I. Luppi, Robin L. Carhart-Harris, Daniel Bor, Anil K. Seth, Adam B. Barrett, 2022https://scholar.google.com/scholar?q=Greater+than+the+parts%3A+a+review+of+the+information+decomposition+approach+to+causal+emergence5. Quantifying High-order Interdependencies via Multivariate Extensions of the Mutual Information — Fernando E. Rosas, Pedro A. M. Mediano, Michael Gastpar, Henrik J. Jensen, 2019https://scholar.google.com/scholar?q=Quantifying+High-order+Interdependencies+via+Multivariate+Extensions+of+the+Mutual+Information6. AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation — Qingyun Wu, Gagan Bansal, Jieyu Zhang, et al., 2023https://scholar.google.com/scholar?q=AutoGen%3A+Enabling+Next-Gen+LLM+Applications+via+Multi-Agent+Conversation7. MetaGPT: Meta Programming for A Multi-Agent Collaborative Framework — Sirui Hong, Mingchen Zhuge, Jonathan Chen, et al., 2024 (preprint 2023)https://scholar.google.com/scholar?q=MetaGPT%3A+Meta+Programming+for+A+Multi-Agent+Collaborative+Framework8. AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors — Weize Chen, Yusheng Su, Jingwei Zuo, et al., 2023https://scholar.google.com/scholar?q=AgentVerse%3A+Facilitating+Multi-Agent+Collaboration+and+Exploring+Emergent+Behaviors9. Evidence for a Collective Intelligence Factor in the Performance of Human Groups — Anita Williams Woolley, Christopher F. Chabris, Alex Pentland, Nada Hashmi, Thomas W. Malone, 2010https://scholar.google.com/scholar?q=Evidence+for+a+Collective+Intelligence+Factor+in+the+Performance+of+Human+Groups10. Machine Theory of Mind — Neil C. Rabinowitz, Frank Perbet, H. Francis Song, Chiyuan Zhang, S. M. Ali Eslami, Matthew Botvinick, 2018https://scholar.google.com/scholar?q=Machine+Theory+of+Mind11. Theory of Mind May Have Spontaneously Emerged in Large Language Models — Michal Kosinski, 2023https://scholar.google.com/scholar?q=Theory+of+Mind+May+Have+Spontaneously+Emerged+in+Large+Language+Models12. Generative Agents: Interactive Simulacra of Human Behavior — Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein, 2023https://scholar.google.com/scholar?q=Generative+Agents%3A+Interactive+Simulacra+of+Human+Behavior13. Human-level play in the game of Diplomacy by combining language models with strategic reasoning (Cicero) — Meta Fundamental AI Research Diplomacy Team (FAIR) et al., 2022https://scholar.google.com/scholar?q=Human-level+play+in+the+game+of+Diplomacy+by+combining+language+models+with+strategic+reasoning+%28Cicero%2914. Williams, P. L., & Beer, R. D. Nonnegative decomposition of multivariate information — Paul L. Williams, Randall D. Beer, 2010https://scholar.google.com/scholar?q=Williams%2C+P.+L.%2C+%26+Beer%2C+R.+D.+Nonnegative+decomposition+of+multivariate+information15. Quantifying unique information — Nils Bertschinger, Johannes Rauh, Eckehard Olbrich, Jürgen Jost, Nihat Ay, 2014https://scholar.google.com/scholar?q=Quantifying+unique+information16. Toward a unified taxonomy of information dynamics via integrated information decomposition (ΦID) — Pedro A. M. Mediano, Fernando E. Rosas, Andrea I. Luppi, et al., 2025https://scholar.google.com/scholar?q=Toward+a+unified+taxonomy+of+information+dynamics+via+integrated+information+decomposition+%28%CE%A6ID%2917. The emergence of specialized roles within groups — Robert L. Goldstone, Edgar J. Andrade-Lotero, Robert D. Hawkins, Michael E. Roberts, 2024https://scholar.google.com/scholar?q=The+emergence+of+specialized+roles+within+groups18. Quantifying collective intelligence in human groups — Christoph Riedl, Young Ji Kim, Pranav Gupta, Thomas W. Malone, Anita Williams Woolley, 2021https://scholar.google.com/scholar?q=Quantifying+collective+intelligence+in+human+groups19. Why do multi-agent LLM systems fail? — Mert Cemri et al., 2025https://scholar.google.com/scholar?q=Why+do+multi-agent+LLM+systems+fail%3F20. Emergent social conventions and collective bias in LLM populations — Ariel Flint Ashery, Luca Maria Aiello, Andrea Baronchelli, 2025https://scholar.google.com/scholar?q=Emergent+social+conventions+and+collective+bias+in+LLM+populations21. Corrupted by reasoning: Reasoning language models become free-riders in public goods games — David Guzman Piedrahita, Yongjin Yang, Mrinmaya Sachan, Giorgia Ramponi, Bernhard Schölkopf, Zhijing Jin, 2025https://scholar.google.com/scholar?q=Corrupted+by+reasoning%3A+Reasoning+language+models+become+free-riders+in+public+goods+games22. Clever Hans or neural theory of mind? Stress testing social reasoning in large language models — Natalie Shapira, Mosh Levy, Seyed Hossein Alavi, Xuhui Zhou, Yejin Choi, Yoav Goldberg, Maarten Sap, Vered Shwartz, 2024https://scholar.google.com/scholar?q=Clever+Hans+or+neural+theory+of+mind%3F+Stress+testing+social+reasoning+in+large+language+models23. More agents is all you need — Junyou Li, Qin Zhang, Yangbin Yu, Qiang Fu, Deheng Ye, 2024https://scholar.google.com/scholar?q=More+agents+is+all+you+needInteractive Visualization: Measuring Emergent Coordination in Multi-Agent LLMs via Information Decomposition17 min
- Emergent Abilities of Large Language Models: Real or Metaphor?This episode examines the 2022 survey "Emergent Abilities of Large Language Models," which argues that some few-shot abilities sit at chance level in small models and then jump sharply in larger ones. It sets up the contrast with smooth scaling laws for cross-entropy loss. It also traces where "emergence" comes from, namely Anderson's "More Is Different" and the loose phase-transition metaphor, and it shows that compute is only a proxy for scale, since data, architecture and training objective all shift where an ability appears. The discussion goes through the evidence: curves stitched together from separately trained models such as GPT-3, LaMDA, Gopher and PaLM, with only a few points per family. Examples include BIG-Bench arithmetic, TruthfulQA, MMLU, and chain-of-thought prompting on GSM8K, which helps only past roughly 10^23 FLOPs. Listeners get a skeptical look at whether the paper actually demonstrates unpredictability, or only builds it into the definition of emergence.Sources:1. Emergent Abilities of Large Language Models — Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed H. Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, William Fedus, 2022http://arxiv.org/abs/2206.076822. Emergent Abilities of Large Language Models — Jason Wei, Yi Tay, Rishi Bommasani, Colin Raffel, Barret Zoph, Sebastian Borgeaud, Dani Yogatama, Maarten Bosma, Denny Zhou, Donald Metzler, Ed Chi, Tatsunori Hashimoto, Oriol Vinyals, Percy Liang, Jeff Dean, William Fedus, 2022https://scholar.google.com/scholar?q=Emergent+Abilities+of+Large+Language+Models3. Are Emergent Abilities of Large Language Models a Mirage? — Rylan Schaeffer, Brando Miranda, Sanmi Koyejo, 2023https://scholar.google.com/scholar?q=Are+Emergent+Abilities+of+Large+Language+Models+a+Mirage%3F4. Predictability and Surprise in Large Generative Models — Deep Ganguli, Danny Hernandez, Liane Lovitt, Amanda Askell, Yuntao Bai, Anna Chen, Tom Conerly, Nova DasSarma, Dawn Drain, Nelson Elhage, Sheer El Showk, Stanislav Fort, Zac Hatfield-Dodds, Tom Henighan, Scott Johnston, Andy Jones, Nicholas Joseph, Jackson Kernian, Shauna Kravec, Ben Mann, Neel Nanda, Kamal Ndousse, Catherine Olsson, Daniela Amodei, Tom Brown, Jared Kaplan, Sam McCandlish, Christopher Olah, Dario Amodei, Jack Clark, 2022https://scholar.google.com/scholar?q=Predictability+and+Surprise+in+Large+Generative+Models5. Beyond the Imitation Game: Quantifying and Extrapolating the Capabilities of Language Models (BIG-bench) — Aarohi Srivastava and 400+ collaborators (including Guy Gur-Ari, Jascha Sohl-Dickstein and Ethan Dyer as core organizers), 2022https://scholar.google.com/scholar?q=Beyond+the+Imitation+Game%3A+Quantifying+and+Extrapolating+the+Capabilities+of+Language+Models+%28BIG-bench%296. More Is Different — Philip W. Anderson, 1972https://scholar.google.com/scholar?q=More+Is+Different7. Scaling Laws for Neural Language Models — Jared Kaplan, Sam McCandlish, Tom Henighan, Tom B. Brown, Benjamin Chess, Rewon Child, Scott Gray, Alec Radford, Jeffrey Wu, Dario Amodei, 2020https://scholar.google.com/scholar?q=Scaling+Laws+for+Neural+Language+Models8. Training Compute-Optimal Large Language Models (Chinchilla) — Jordan Hoffmann, Sebastian Borgeaud, Arthur Mensch, Elena Buchatskaya, Trevor Cai, Eliza Rutherford, Diego de Las Casas, Lisa Anne Hendricks, Johannes Welbl, Aidan Clark, Tom Hennigan, Eric Noland, Katie Millican, George van den Driessche, Bogdan Damoc, Aurelia Guy, Simon Osindero, Karen Simonyan, Erich Elsen, Jack W. Rae, Oriol Vinyals, Laurent Sifre, 2022https://scholar.google.com/scholar?q=Training+Compute-Optimal+Large+Language+Models+%28Chinchilla%299. Grokking: Generalization Beyond Overfitting on Small Algorithmic Datasets — Alethea Power, Yuri Burda, Harri Edwards, Igor Babuschkin, Vedant Misra, 2022https://scholar.google.com/scholar?q=Grokking%3A+Generalization+Beyond+Overfitting+on+Small+Algorithmic+Datasets10. In-context Learning and Induction Heads — Catherine Olsson et al., 2022https://scholar.google.com/scholar?q=In-context+Learning+and+Induction+Heads11. Progress Measures for Grokking via Mechanistic Interpretability — Neel Nanda, Lawrence Chan, Tom Lieberum, Jess Smith, Jacob Steinhardt, 2023https://scholar.google.com/scholar?q=Progress+Measures+for+Grokking+via+Mechanistic+Interpretability12. Chain-of-Thought Prompting Elicits Reasoning in Large Language Models — Jason Wei et al., 2022https://scholar.google.com/scholar?q=Chain-of-Thought+Prompting+Elicits+Reasoning+in+Large+Language+Models13. Finetuned Language Models Are Zero-Shot Learners (FLAN) — Jason Wei et al., 2022https://scholar.google.com/scholar?q=Finetuned+Language+Models+Are+Zero-Shot+Learners+%28FLAN%2914. Show Your Work: Scratchpads for Intermediate Computation with Language Models — Maxwell Nye et al., 2021https://scholar.google.com/scholar?q=Show+Your+Work%3A+Scratchpads+for+Intermediate+Computation+with+Language+Models15. Language Models (Mostly) Know What They Know — Saurav Kadavath et al., 2022https://scholar.google.com/scholar?q=Language+Models+%28Mostly%29+Know+What+They+Know16. Inverse Scaling Prize / Inverse Scaling: When Bigger Isn't Better — Ian McKenzie et al., 2022–2023https://scholar.google.com/scholar?q=Inverse+Scaling+Prize+%2F+Inverse+Scaling%3A+When+Bigger+Isn%27t+Better17. Broken Neural Scaling Laws — Ethan Caballero, Kshitij Gupta, Irina Rish, David Krueger, 2022https://scholar.google.com/scholar?q=Broken+Neural+Scaling+Laws18. Transcending Scaling Laws with 0.1% Extra Compute (UL2R) — Yi Tay et al., 2022https://scholar.google.com/scholar?q=Transcending+Scaling+Laws+with+0.1%25+Extra+Compute+%28UL2R%2919. Understanding Emergent Abilities of Language Models from the Loss Perspective — Zhengxiao Du et al., 2024https://scholar.google.com/scholar?q=Understanding+Emergent+Abilities+of+Language+Models+from+the+Loss+Perspective20. Data Distributional Properties Drive Emergent In-Context Learning in Transformers — Stephanie Chan et al., 2022https://scholar.google.com/scholar?q=Data+Distributional+Properties+Drive+Emergent+In-Context+Learning+in+TransformersInteractive Visualization: Emergent Abilities of Large Language Models: Real or Metaphor?17 min
- LoCoMo Benchmark: Do LLM Agents Remember Very Long Conversations?This episode examines LoCoMo, a benchmark for testing whether LLM agents remember across very long conversations. The conversations average about 9,200 tokens, 19 sessions and 305 turns, roughly nine times longer than earlier datasets such as Multi-Session Chat. The episode covers how the conversations are generated: two LLM agents get personas and a temporal event graph of up to 25 causally linked life events. Human annotators then edited about 15 percent of turns, which raises questions about how much of the data was verified. It compares three ways a model can handle history: truncated context, long-context models, and retrieval-augmented generation. It also walks through the question-answering categories (single-hop, multi-hop, temporal, open-domain, adversarial) and F1 scoring. The paper's headline finding is that every approach falls well short of human performance. The episode also previews how later memory products such as MemGPT, Mem0, and Zep use the benchmark, and the leaderboard disputes that followed.Sources:1. Evaluating Very Long-Term Conversational Memory of LLM Agents — Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, Yuwei Fang, 2024http://arxiv.org/abs/2402.177532. Evaluating Very Long-Term Conversational Memory of LLM Agents (LoCoMo) — Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, Yuwei Fang, 2024 (ACL 2024)https://scholar.google.com/scholar?q=Evaluating+Very+Long-Term+Conversational+Memory+of+LLM+Agents+%28LoCoMo%293. Recursively Summarizing Enables Long-Term Dialogue Memory in Large Language Models — Qingyue Wang, Liang Ding, Yanan Cao, Zhiliang Tian, Shi Wang, Dacheng Tao, Li Guo, 2023https://scholar.google.com/scholar?q=Recursively+Summarizing+Enables+Long-Term+Dialogue+Memory+in+Large+Language+Models4. MemGPT: Towards LLMs as Operating Systems — Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G. Patil, Ion Stoica, Joseph E. Gonzalez, 2023https://scholar.google.com/scholar?q=MemGPT%3A+Towards+LLMs+as+Operating+Systems5. Generative Agents: Interactive Simulacra of Human Behavior — Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein, 2023https://scholar.google.com/scholar?q=Generative+Agents%3A+Interactive+Simulacra+of+Human+Behavior6. Beyond Goldfish Memory: Long-Term Open-Domain Conversation (Multi-Session Chat) — Jing Xu, Arthur Szlam, Jason Weston, 2022 (ACL 2022)https://scholar.google.com/scholar?q=Beyond+Goldfish+Memory%3A+Long-Term+Open-Domain+Conversation+%28Multi-Session+Chat%297. MMDialog: A Large-scale Multi-turn Dialogue Dataset Towards Multi-modal Open-domain Conversation — Jiazhan Feng, Qingfeng Sun, Can Xu, Pu Zhao, Yaming Yang, Chongyang Tao, Dongyan Zhao, Qingwei Lin, 2023 (ACL 2023)https://scholar.google.com/scholar?q=MMDialog%3A+A+Large-scale+Multi-turn+Dialogue+Dataset+Towards+Multi-modal+Open-domain+Conversation8. PhotoChat: A Human-Human Dialogue Dataset with Photo Sharing Behavior for Joint Image-Text Modeling — Xiaoxue Zang, Lijuan Liu, Maria Wang, Yang Song, Hao Zhang, Jindong Chen, 2021 (ACL 2021)https://scholar.google.com/scholar?q=PhotoChat%3A+A+Human-Human+Dialogue+Dataset+with+Photo+Sharing+Behavior+for+Joint+Image-Text+Modeling9. Visual Instruction Tuning (LLaVA) — Haotian Liu, Chunyuan Li, Qingyang Wu, Yong Jae Lee, 2023 (NeurIPS 2023)https://scholar.google.com/scholar?q=Visual+Instruction+Tuning+%28LLaVA%2910. BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models — Junnan Li, Dongxu Li, Silvio Savarese, Steven Hoi, 2023 (ICML 2023)https://scholar.google.com/scholar?q=BLIP-2%3A+Bootstrapping+Language-Image+Pre-training+with+Frozen+Image+Encoders+and+Large+Language+Models11. Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory — Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, Deshraj Yadav, 2025https://scholar.google.com/scholar?q=Mem0%3A+Building+Production-Ready+AI+Agents+with+Scalable+Long-Term+Memory12. Zep: A Temporal Knowledge Graph Architecture for Agent Memory — Preston Rasmussen, Pavlo Paliychuk, Travis Beauvais, Jack Ryan, Daniel Chalef, 2025https://scholar.google.com/scholar?q=Zep%3A+A+Temporal+Knowledge+Graph+Architecture+for+Agent+Memory13. A-Mem: Agentic Memory for LLM Agents — Wujiang Xu et al., 2025https://scholar.google.com/scholar?q=A-Mem%3A+Agentic+Memory+for+LLM+Agents14. LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory — Di Wu, Hongwei Wang, Wenhao Yu, Yuwei Zhang, Kai-Wei Chang, Dong Yu, 2024https://scholar.google.com/scholar?q=LongMemEval%3A+Benchmarking+Chat+Assistants+on+Long-Term+Interactive+Memory15. Lost in the Middle: How Language Models Use Long Contexts — Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, Percy Liang, 2024https://scholar.google.com/scholar?q=Lost+in+the+Middle%3A+How+Language+Models+Use+Long+Contexts16. Beyond Goldfish Memory: Long-Term Open-Domain Conversation (MSC) — Jing Xu, Arthur Szlam, Jason Weston, 2022https://scholar.google.com/scholar?q=Beyond+Goldfish+Memory%3A+Long-Term+Open-Domain+Conversation+%28MSC%2917. Titans: Learning to Memorize at Test Time — Ali Behrouz, Peilin Zhong, Vahab Mirrokni, 2024https://scholar.google.com/scholar?q=Titans%3A+Learning+to+Memorize+at+Test+Time18. Needle-in-a-haystack and RULER: What's the Real Context Size of Your Long-Context Language Models? — Cheng-Ping Hsieh et al., 2024https://scholar.google.com/scholar?q=Needle-in-a-haystack+and+RULER%3A+What%27s+the+Real+Context+Size+of+Your+Long-Context+Language+Models%3FInteractive Visualization: LoCoMo Benchmark: Do LLM Agents Remember Very Long Conversations?0 min
- MAGMA: Multi-Graph Agentic Memory for Long-Horizon AI AgentsThis episode examines MAGMA, a memory architecture for long-running AI agents that stores each event once and links it across four separate graphs: semantic, temporal, causal, and entity. Retrieval follows whichever edge type a question calls for. The discussion first sets out why bigger context windows don't solve the problem, then traces the shift from long-context models to RAG to agent-owned memory systems such as MemGPT, A-Mem, and Zep. It argues that these systems organize memory by similarity rather than by dependency, so they struggle with "why" and "what came before" questions. The hosts also describe the LoCoMo and LongMemEval benchmarks, point out that MAGMA is a data structure around a frozen LLM rather than a trained network, and note that the extra write-time LLM extraction (especially for causal edges) raises the question of whether the headline gains come from the graph or from the surrounding prompts. Listeners get a skeptical look at what the paper's single headline number actually measures, and at which comparisons, such as Mem0 and HippoRAG 2, the paper cites but doesn't run.Sources:1. MAGMA: A Multi-Graph based Agentic Memory Architecture for AI Agents — Dongming Jiang, Yi Li, Guanpeng Li, Bingzhe Li, 2026http://arxiv.org/abs/2601.032362. MemGPT: Towards LLMs as Operating Systems — Charles Packer, Sarah Wooders, Kevin Lin, Vivian Fang, Shishir G. Patil, Ion Stoica, Joseph E. Gonzalez, 2023https://scholar.google.com/scholar?q=MemGPT%3A+Towards+LLMs+as+Operating+Systems3. Generative Agents: Interactive Simulacra of Human Behavior — Joon Sung Park, Joseph C. O'Brien, Carrie J. Cai, Meredith Ringel Morris, Percy Liang, Michael S. Bernstein, 2023https://scholar.google.com/scholar?q=Generative+Agents%3A+Interactive+Simulacra+of+Human+Behavior4. Zep: A Temporal Knowledge Graph Architecture for Agent Memory — Preston Rasmussen, Pavlo Paliychuk, Travis Beauvais, Jack Ryan, Daniel Chalef, 2025https://scholar.google.com/scholar?q=Zep%3A+A+Temporal+Knowledge+Graph+Architecture+for+Agent+Memory5. A-Mem: Agentic Memory for LLM Agents — Wujiang Xu, Zujie Liang, Kai Mei, Hang Gao, Juntao Tan, Yongfeng Zhang, 2025https://scholar.google.com/scholar?q=A-Mem%3A+Agentic+Memory+for+LLM+Agents6. Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory — Prateek Chhikara, Dev Khant, Saket Aryan, Taranjeet Singh, Deshraj Yadav, 2025https://scholar.google.com/scholar?q=Mem0%3A+Building+Production-Ready+AI+Agents+with+Scalable+Long-Term+Memory7. Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks — Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Douwe Kiela and others, 2020https://scholar.google.com/scholar?q=Retrieval-Augmented+Generation+for+Knowledge-Intensive+NLP+Tasks8. Evaluating Very Long-Term Conversational Memory of LLM Agents (LoCoMo) — Adyasha Maharana, Dong-Ho Lee, Sergey Tulyakov, Mohit Bansal, Francesco Barbieri, Yuwei Fang, 2024https://scholar.google.com/scholar?q=Evaluating+Very+Long-Term+Conversational+Memory+of+LLM+Agents+%28LoCoMo%299. LongMemEval: Benchmarking Chat Assistants on Long-Term Interactive Memory — Di Wu, Hongwei Wang, Wenhao Yu, Yuwei Zhang, Kai-Wei Chang, Dong Yu, 2024https://scholar.google.com/scholar?q=LongMemEval%3A+Benchmarking+Chat+Assistants+on+Long-Term+Interactive+Memory10. A Survey on the Memory Mechanism of Large Language Model based Agents — Zeyu Zhang, Xiaohe Bo, Chen Ma, Rui Li, Xu Chen and others, 2024https://scholar.google.com/scholar?q=A+Survey+on+the+Memory+Mechanism+of+Large+Language+Model+based+Agents11. From Local to Global: A Graph RAG Approach to Query-Focused Summarization — Darren Edge, Ha Trinh, Newman Cheng, Joshua Bradley, Alex Chao, Apurva Mody, Steven Truitt, Jonathan Larson, 2024https://scholar.google.com/scholar?q=From+Local+to+Global%3A+A+Graph+RAG+Approach+to+Query-Focused+Summarization12. HippoRAG: Neurobiologically Inspired Long-Term Memory for Large Language Models — Bernal Jiménez Gutiérrez, Yiheng Shu, Yu Gu, Michihiro Yasunaga, Yu Su, 2024https://scholar.google.com/scholar?q=HippoRAG%3A+Neurobiologically+Inspired+Long-Term+Memory+for+Large+Language+Models13. Think-on-Graph: Deep and Responsible Reasoning of Large Language Model on Knowledge Graph — Jiashuo Sun, Chengjin Xu, Lumingyi Tang, Saizhuo Wang, Chen Lin, Yeyun Gong, Lionel M. Ni, Heung-Yeung Shum, Jian Guo, 2023https://scholar.google.com/scholar?q=Think-on-Graph%3A+Deep+and+Responsible+Reasoning+of+Large+Language+Model+on+Knowledge+Graph14. Go for a Walk and Arrive at the Answer: Reasoning Over Paths in Knowledge Bases using Reinforcement Learning (MINERVA) — Rajarshi Das, Shehzaad Dhuliawala, Manzil Zaheer, Luke Vilnis, Ishan Durugkar, Akshay Krishnamurthy, Alex Smola, Andrew McCallum, 2018https://scholar.google.com/scholar?q=Go+for+a+Walk+and+Arrive+at+the+Answer%3A+Reasoning+Over+Paths+in+Knowledge+Bases+using+Reinforcement+Learning+%28MINERVA%2915. Mem0: Building Production-Ready AI Agents with Scalable Long-Term Memory (including Mem0^g graph variant) — Chhikara, Khant, Aryan, Singh, Yadav, 2025https://scholar.google.com/scholar?q=Mem0%3A+Building+Production-Ready+AI+Agents+with+Scalable+Long-Term+Memory+%28including+Mem0%5Eg+graph+variant%2916. HippoRAG 2 / From RAG to Memory: Non-Parametric Continual Learning for LLMs — Gutiérrez, Shu, Qi, Zhou, Su, 2025https://scholar.google.com/scholar?q=HippoRAG+2+%2F+From+RAG+to+Memory%3A+Non-Parametric+Continual+Learning+for+LLMs17. Nemori: Self-Organizing Agent Memory Inspired by Cognitive Science — Nan, Ma, Wu, Chen, 2025https://scholar.google.com/scholar?q=Nemori%3A+Self-Organizing+Agent+Memory+Inspired+by+Cognitive+Science18. Anatomy of Agentic Memory: Taxonomy and Empirical Analysis of Evaluation and System Limitations — Jiang, Li, Wei, et al., 2026https://scholar.google.com/scholar?q=Anatomy+of+Agentic+Memory%3A+Taxonomy+and+Empirical+Analysis+of+Evaluation+and+System+Limitations19. Memory in the Age of AI Agents (survey) — Hu et al., 2025https://scholar.google.com/scholar?q=Memory+in+the+Age+of+AI+Agents+%28survey%2920. Think-on-Graph: Deep and Responsible Reasoning of LLMs on Knowledge Graph — Sun et al., 2024https://scholar.google.com/scholar?q=Think-on-Graph%3A+Deep+and+Responsible+Reasoning+of+LLMs+on+Knowledge+Graph21. Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena — Zheng et al., 2023https://scholar.google.com/scholar?q=Judging+LLM-as-a-Judge+with+MT-Bench+and+Chatbot+ArenaInteractive Visualization: MAGMA: Multi-Graph Agentic Memory for Long-Horizon AI Agents19 min
- Chunk-Aware KV Cache Management Bridges LLM Serving and CXL-SSDsThis episode examines why a stock CXL-attached SSD, meaning flash capacity exposed as load/store memory behind a small device-side DRAM cache, serves reused prompt prefixes about three times slower than local DRAM. It explains how prefix caching cuts time-to-first-token by skipping prefill, and how LMCache stores KV in fixed-size token chunks across GPU, host DRAM, and storage tiers. A ramdisk experiment on Qwen3-4B with a 34k-token prefix shows the block I/O interface alone costs 1.8x in TTFT, through staging copies and last-level cache contention, even with no flash involved. A generic next-n prefetcher in the device also fails, because the device sees only page addresses while the serving engine knows which chunks a request needs and in what order. That semantic gap motivates the paper's redesign around the chunk as the shared unit between engine and device, and the episode compares it briefly with the concurrent HyMCache and ITME work.Sources:1. Bridging LLM Serving and CXL-SSDs with Chunk-Aware KV Cache Management — Hyunsun Chung, Taewan Noh, Minji Kim, Joo-Young Hwang, Hong-Yeon Kim, Youngjae Kim, 2026http://arxiv.org/abs/2609.26828v12. Pond: CXL-Based Memory Pooling Systems for Cloud Platforms — Huaicheng Li, Daniel S. Berger, Lisa Hsu, Daniel Ernst, Pantea Zardoshti, Stanko Novakovic, Monish Shah, Samir Rajadnya, Scott Lee, Ishwar Agarwal, Mark D. Hill, Marcus Fontoura, Ricardo Bianchini, 2023https://scholar.google.com/scholar?q=Pond%3A+CXL-Based+Memory+Pooling+Systems+for+Cloud+Platforms3. TPP: Transparent Page Placement for CXL-Enabled Tiered-Memory — Hasan Al Maruf, Hao Wang, Abhishek Dhanotia, Johannes Weiner, Niket Agarwal, Pallab Bhattacharya, Chris Petersen, Mosharaf Chowdhury, Shobhit Kanaujia, Prakash Chauhan, 2023https://scholar.google.com/scholar?q=TPP%3A+Transparent+Page+Placement+for+CXL-Enabled+Tiered-Memory4. Efficient Memory Management for Large Language Model Serving with PagedAttention — Woosuk Kwon, Zhuohan Li, Siyuan Zhuang, Ying Sheng, Lianmin Zheng, Cody Hao Yu, Joseph Gonzalez, Hao Zhang, Ion Stoica, 2023https://scholar.google.com/scholar?q=Efficient+Memory+Management+for+Large+Language+Model+Serving+with+PagedAttention5. CacheBlend: Fast Large Language Model Serving for RAG with Cached Knowledge Fusion, and the LMCache system line (LMCache: An Efficient KV Cache Layer for Enterprise-Scale LLM Inference) — Jiayi Yao, Hanchen Li, Yuhan Liu, Siddhant Ray, Yihua Cheng, Qizheng Zhang, Kuntai Du, Shan Lu, Junchen Jiang (CacheBlend); LMCache team from the University of Chicago and Tensormesh (LMCache), 2024-2025https://scholar.google.com/scholar?q=CacheBlend%3A+Fast+Large+Language+Model+Serving+for+RAG+with+Cached+Knowledge+Fusion%2C+and+the+LMCache+system+line+%28LMCache%3A+An+Efficient+KV+Cache+Layer+for+Enterprise-Scale+LLM+Inference%296. HyMCache: A KV cache framework for multi-turn LLM serving with CXL-Hybrid Memory — Hakbeom Jang, Inho Song, Hoshik Kim, Sam H. Noh, Jongryool Kim, 2026https://scholar.google.com/scholar?q=HyMCache%3A+A+KV+cache+framework+for+multi-turn+LLM+serving+with+CXL-Hybrid+Memory7. ITME: Inference tiered memory expansion with disaggregated CXL-hybrid memories — Hakbeom Jang et al., 2026https://scholar.google.com/scholar?q=ITME%3A+Inference+tiered+memory+expansion+with+disaggregated+CXL-hybrid+memories8. Cylon: Fast and accurate full-system emulation of CXL-SSDs — Dongha Yoon, Hansen Idden, Jinshu Liu, Berkay Inceisci, Sam H. Noh, Huaicheng Li, 2026https://scholar.google.com/scholar?q=Cylon%3A+Fast+and+accurate+full-system+emulation+of+CXL-SSDs9. SkyByte: Architecting an efficient memory-semantic CXL-based SSD with OS and hardware co-design — Haoyang Zhang, Yuqi Xue, Yirui Eric Zhou, Shaobo Li, Jian Huang, 2025https://scholar.google.com/scholar?q=SkyByte%3A+Architecting+an+efficient+memory-semantic+CXL-based+SSD+with+OS+and+hardware+co-design10. Overcoming the memory wall with CXL-Enabled SSDs — Shao-Peng Yang, Minjae Kim, Sanghyun Nam, Juhyung Park, Jin-Yong Choi, Eyee Hyun Nam, Eunji Lee, Sungjin Lee, Bryan S. Kim, 2023https://scholar.google.com/scholar?q=Overcoming+the+memory+wall+with+CXL-Enabled+SSDs11. Performance characterizations and usage guidelines of Samsung CXL Memory Module Hybrid prototype — Jianping Zeng et al., 2025https://scholar.google.com/scholar?q=Performance+characterizations+and+usage+guidelines+of+Samsung+CXL+Memory+Module+Hybrid+prototype12. CacheGen: KV cache compression and streaming for fast large language model serving — Yuhan Liu, Hanchen Li, Yihua Cheng et al., 2024https://scholar.google.com/scholar?q=CacheGen%3A+KV+cache+compression+and+streaming+for+fast+large+language+model+serving13. Mooncake: A KVCache-centric disaggregated architecture for LLM serving — Ruoyu Qin et al., 2025https://scholar.google.com/scholar?q=Mooncake%3A+A+KVCache-centric+disaggregated+architecture+for+LLM+serving14. DualPath: Accelerating agentic LLM inference by harvesting disaggregated KV-cache storage I/O — Yongtong Wu et al., 2026https://scholar.google.com/scholar?q=DualPath%3A+Accelerating+agentic+LLM+inference+by+harvesting+disaggregated+KV-cache+storage+I%2FO15. CXL topology-aware and expander-driven prefetching: Unlocking solid-state drive performance — Dongsuk Oh et al., 2025https://scholar.google.com/scholar?q=CXL+topology-aware+and+expander-driven+prefetching%3A+Unlocking+solid-state+drive+performance0 min
- BABILong: Testing LLM Long-Context Reasoning in a HaystackThis episode examines BABILong, a NeurIPS 2024 benchmark that asks how much of a language model's advertised 128K-plus context window it can use for reasoning. It argues that the popular needle-in-a-haystack test mostly measures lookup, which strong models pass almost perfectly. BABILong hides the sentences of the synthetic bAbI reasoning tasks (chaining facts, counting, temporal ordering) inside long passages of PG19 book text. Answering then requires finding and combining several scattered facts, and the benchmark can be generated at any length, up to millions of tokens, without risk of training-set contamination. The discussion defines "effective context size" as the longest input at which accuracy stays above 85%, and it flags that this threshold is a judgment call. It also sets up comparisons between long-context LLMs, retrieval-augmented generation, and small recurrent-memory and state-space models. The hosts note that some of the winning models are trained on the task and others are not, so listeners get both the benchmark's design and a caution about reading its results.Sources:1. BABILong: Testing the Limits of LLMs with Long Context Reasoning-in-a-Haystack — Yuri Kuratov, Aydar Bulatov, Petr Anokhin, Ivan Rodkin, Dmitry Sorokin, Artyom Sorokin, Mikhail Burtsev, 2024http://arxiv.org/abs/2406.101492. Towards AI-Complete Question Answering: A Set of Prerequisite Toy Tasks (bAbI) — Jason Weston, Antoine Bordes, Sumit Chopra, Alexander M. Rush, Bart van Merriënboer, Armand Joulin, Tomas Mikolov, 2015 (ICLR 2016)https://scholar.google.com/scholar?q=Towards+AI-Complete+Question+Answering%3A+A+Set+of+Prerequisite+Toy+Tasks+%28bAbI%293. Lost in the Middle: How Language Models Use Long Contexts — Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, Percy Liang, 2023 (TACL 2024)https://scholar.google.com/scholar?q=Lost+in+the+Middle%3A+How+Language+Models+Use+Long+Contexts4. RULER: What's the Real Context Size of Your Long-Context Language Models? — Cheng-Ping Hsieh, Simeng Sun, Samuel Kriman, Shantanu Acharya, Dima Rekesh, Fei Jia, Yang Zhang, Boris Ginsburg (NVIDIA), 2024 (COLM 2024)https://scholar.google.com/scholar?q=RULER%3A+What%27s+the+Real+Context+Size+of+Your+Long-Context+Language+Models%3F5. NoLiMa: Long-Context Evaluation Beyond Literal Matching — Ali Modarressi, Hanieh Deilamsalehy, Franck Dernoncourt, Trung Bui, Ryan A. Rossi, Seunghyun Yoon, Hinrich Schütze (Adobe Research / LMU), 2025https://scholar.google.com/scholar?q=NoLiMa%3A+Long-Context+Evaluation+Beyond+Literal+Matching6. Recurrent Memory Transformer — Aydar Bulatov, Yuri Kuratov, Mikhail Burtsev, 2022 (NeurIPS 2022)https://scholar.google.com/scholar?q=Recurrent+Memory+Transformer7. Scaling Transformer to 1M tokens and beyond with RMT — Aydar Bulatov, Yuri Kuratov, Yermek Kapushev, Mikhail S. Burtsev, 2023https://scholar.google.com/scholar?q=Scaling+Transformer+to+1M+tokens+and+beyond+with+RMT8. Associative Recurrent Memory Transformer (ARMT) — Ivan Rodkin, Yuri Kuratov, Aydar Bulatov, Mikhail Burtsev, 2024https://scholar.google.com/scholar?q=Associative+Recurrent+Memory+Transformer+%28ARMT%299. Leave No Context Behind: Efficient Infinite Context Transformers with Infini-attention — Tsendsuren Munkhdalai, Manaal Faruqui, Siddharth Gopal (Google), 2024https://scholar.google.com/scholar?q=Leave+No+Context+Behind%3A+Efficient+Infinite+Context+Transformers+with+Infini-attention10. Associative Recurrent Memory Transformer — Rodkin, Kuratov, Bulatov, Burtsev, 2024https://scholar.google.com/scholar?q=Associative+Recurrent+Memory+Transformer11. Scaling Transformer to 1M tokens and beyond with RMT / Beyond Attention: Breaking the Limits of Transformer Context Length with Recurrent Memory — Bulatov, Kuratov, Kapushev, Burtsev, 2023/2024https://scholar.google.com/scholar?q=Scaling+Transformer+to+1M+tokens+and+beyond+with+RMT+%2F+Beyond+Attention%3A+Breaking+the+Limits+of+Transformer+Context+Length+with+Recurrent+Memory12. Gated Delta Networks: Improving Mamba2 with Delta Rule — Yang, Kautz, Hatamizadeh, 2024https://scholar.google.com/scholar?q=Gated+Delta+Networks%3A+Improving+Mamba2+with+Delta+Rule13. Titans: Learning to Memorize at Test Time — Behrouz, Zhong, Mirrokni, 2024/2025https://scholar.google.com/scholar?q=Titans%3A+Learning+to+Memorize+at+Test+Time14. Transformer-XL and Compressive Transformer (PG19) — Dai et al. 2019; Rae et al. 2020, 2019/2020https://scholar.google.com/scholar?q=Transformer-XL+and+Compressive+Transformer+%28PG19%2915. Infini-attention (Leave No Context Behind: Efficient Infinite Context Transformers with Infini-attention) — Munkhdalai, Faruqui, Gopal, 2024https://scholar.google.com/scholar?q=Infini-attention+%28Leave+No+Context+Behind%3A+Efficient+Infinite+Context+Transformers+with+Infini-attention%2916. Counting-Stars and InfiniteBench — Song et al.; Zhang et al., 2024https://scholar.google.com/scholar?q=Counting-Stars+and+InfiniteBenchInteractive Visualization: BABILong: Testing LLM Long-Context Reasoning in a Haystack17 min
- Query-Key Normalization Improves Low-Resource Transformer TranslationThis episode examines Query-Key Normalization for Transformers, which swaps the attention dot product for a scaled cosine similarity. It reports an average gain of 0.928 BLEU across five low-resource translation pairs, using small models and corpora of roughly 10,000 to 214,000 sentence pairs. The discussion explains why unbounded dot products can push softmax toward near one-hot attention, and how L2-normalizing queries and keys bounds the scores. A learnable scale, initialized by a rule of thumb based on sequence length, then restores attention sharpness. It also places the method against the ScaleNorm, FixNorm and PreNorm baseline, and flags a confound: the baseline uses ScaleNorm while the QKNorm model uses LayerNorm. Listeners get a careful look at how much of the gain the change itself can claim.Sources:1. Query-Key Normalization for Transformers — Alex Henry, Prudhvi Raj Dachapally, Shubham Pawar, Yuxuan Chen, 2020http://arxiv.org/abs/2010.042452. Transformers without Tears: Improving the Normalization of Self-Attention — Toan Q. Nguyen, Julian Salazar, 2019https://scholar.google.com/scholar?q=Transformers+without+Tears%3A+Improving+the+Normalization+of+Self-Attention3. Query-Key Normalization for Transformers — Alex Henry, Prudhvi Raj Dachapally, Shubham Pawar, Yuxuan Chen, 2020https://scholar.google.com/scholar?q=Query-Key+Normalization+for+Transformers4. Scaling Vision Transformers to 22 Billion Parameters — Mostafa Dehghani, Josip Djolonga, Basil Mustafa, et al. (Google Research), 2023https://scholar.google.com/scholar?q=Scaling+Vision+Transformers+to+22+Billion+Parameters5. Small-scale proxies for large-scale Transformer training instabilities — Mitchell Wortsman, Peter J. Liu, Lechao Xiao, et al., 2023https://scholar.google.com/scholar?q=Small-scale+proxies+for+large-scale+Transformer+training+instabilities6. Revisiting Low-Resource Neural Machine Translation: A Case Study — Rico Sennrich, Biao Zhang, 2019https://scholar.google.com/scholar?q=Revisiting+Low-Resource+Neural+Machine+Translation%3A+A+Case+Study7. Massively Multilingual Neural Machine Translation — Roee Aharoni, Melvin Johnson, Orhan Firat, 2019https://scholar.google.com/scholar?q=Massively+Multilingual+Neural+Machine+Translation8. When and Why Are Pre-trained Word Embeddings Useful for Neural Machine Translation? — Ye Qi, Devendra Singh Sachan, Matthieu Felix, Sarguna Padmanabhan, Graham Neubig, 2018https://scholar.google.com/scholar?q=When+and+Why+Are+Pre-trained+Word+Embeddings+Useful+for+Neural+Machine+Translation%3F9. No Language Left Behind: Scaling Human-Centered Machine Translation — NLLB Team (Marta R. Costa-jussà et al., Meta AI), 2022https://scholar.google.com/scholar?q=No+Language+Left+Behind%3A+Scaling+Human-Centered+Machine+Translation10. Query-Key Normalization for Transformers (follow-up usage) / Cosine-attention in Swin Transformer V2 — Ze Liu et al., 2022https://scholar.google.com/scholar?q=Query-Key+Normalization+for+Transformers+%28follow-up+usage%29+%2F+Cosine-attention+in+Swin+Transformer+V211. On Layer Normalization in the Transformer Architecture — Ruibin Xiong et al., 2020https://scholar.google.com/scholar?q=On+Layer+Normalization+in+the+Transformer+Architecture12. Normalized Attention Without Probability Cage — Oliver Richter, Roger Wattenhofer, 2020https://scholar.google.com/scholar?q=Normalized+Attention+Without+Probability+Cage13. Attention Is Not All You Need / On the Expressivity Role of LayerNorm and attention entropy collapse (Zhai et al., 'Stabilizing Transformer Training by Preventing Attention Entropy Collapse') — Shuangfei Zhai et al., 2023https://scholar.google.com/scholar?q=Attention+Is+Not+All+You+Need+%2F+On+the+Expressivity+Role+of+LayerNorm+and+attention+entropy+collapse+%28Zhai+et+al.%2C+%27Stabilizing+Transformer+Training+by+Preventing+Attention+Entropy+Collapse%27%2914. Root Mean Square Layer Normalization — Biao Zhang, Rico Sennrich, 2019https://scholar.google.com/scholar?q=Root+Mean+Square+Layer+Normalization15. BPE-Dropout: Simple and Effective Subword Regularization — Ivan Provilkov, Dmitrii Emelianenko, Elena Voita, 2020https://scholar.google.com/scholar?q=BPE-Dropout%3A+Simple+and+Effective+Subword+RegularizationInteractive Visualization: Query-Key Normalization Improves Low-Resource Transformer Translation16 min
About AI Post Transformers
From the publisher's feed
AI-generated podcast where hosts Hal Turing and Dr. Ada Shannon discuss the latest research papers and reports in machine learning, AI systems, and optimization. Featuring honest critical analysis,…
proper citations, and nerdy humor.