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
- Transformer Layers as Painters: Skipping and Reordering Frozen LLM LayersThis episode examines "Transformer Layers as Painters," which asks whether a frozen, ordinarily trained Llama 2 can tolerate having its layers skipped, swapped, reordered, or run in parallel without any retraining. The discussion uses a painter-on-an-assembly-line analogy: the residual stream serves as a shared canvas, so layers read and write the same space. It places the paper alongside related work on residual networks, logit and tuned lens, depth pruning, and the "stages of inference" study. Testing on Llama2-7B, 13B, and 70B plus BERT-Large across five benchmarks, the first results show a sharp split. Removing or swapping the first and last layers collapses performance, while the uniform middle layers barely register a change, and accuracy falls off gradually instead of breaking. Listeners interested in layer pruning, conditional computation, and the latency cost of depth get a look at what a pretrained model can absorb without being trained for it.Sources:1. Transformer Layers as Painters — Qi Sun, Marc Pickett, Aakash Kumar Nain, Llion Jones, 2024http://arxiv.org/abs/2407.092982. Eliciting Latent Predictions from Transformers with the Tuned Lens — Nora Belrose, Zach Furman, Logan Smith, Danny Halawi, Igor Ostrovsky, Lev McKinney, Stella Biderman, Jacob Steinhardt (building on nostalgebraist's 2020 'interpreting GPT: the logit lens'), 2023https://scholar.google.com/scholar?q=Eliciting+Latent+Predictions+from+Transformers+with+the+Tuned+Lens3. The Remarkable Robustness of LLMs: Stages of Inference? — Vedang Lad, Wes Gurnee, Max Tegmark, 2024https://scholar.google.com/scholar?q=The+Remarkable+Robustness+of+LLMs%3A+Stages+of+Inference%3F4. The Unreasonable Ineffectiveness of the Deeper Layers — Andrey Gromov, Kushal Tirumala, Hassan Shapourian, Paolo Glorioso, Daniel A. Roberts, 2024https://scholar.google.com/scholar?q=The+Unreasonable+Ineffectiveness+of+the+Deeper+Layers5. Residual Networks Behave Like Ensembles of Relatively Shallow Networks — Andreas Veit, Michael Wilber, Serge Belongie, 2016https://scholar.google.com/scholar?q=Residual+Networks+Behave+Like+Ensembles+of+Relatively+Shallow+Networks6. Reducing Transformer Depth on Demand with Structured Dropout (LayerDrop) — Angela Fan, Edouard Grave, Armand Joulin, 2019 (ICLR 2020)https://scholar.google.com/scholar?q=Reducing+Transformer+Depth+on+Demand+with+Structured+Dropout+%28LayerDrop%297. LayerSkip: Enabling Early Exit Inference and Self-Speculative Decoding — Mostafa Elhoushi, Akshat Shrivastava, Diana Liskovich, Basil Hosmer, Bram Wasti, et al. (Meta), 2024https://scholar.google.com/scholar?q=LayerSkip%3A+Enabling+Early+Exit+Inference+and+Self-Speculative+Decoding8. Mixture-of-Depths: Dynamically Allocating Compute in Transformer-Based Language Models — David Raposo, Sam Ritter, Blake Richards, Timothy Lillicrap, Peter Conway Humphreys, Adam Santoro (Google DeepMind), 2024https://scholar.google.com/scholar?q=Mixture-of-Depths%3A+Dynamically+Allocating+Compute+in+Transformer-Based+Language+Models9. Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach — Jonas Geiping, Sean McLeish, Neel Jain, John Kirchenbauer, Siddharth Singh, Brian R. Bartoldson, Bhavya Kailkhura, Abhinav Bhatele, Tom Goldstein, 2025https://scholar.google.com/scholar?q=Scaling+up+Test-Time+Compute+with+Latent+Reasoning%3A+A+Recurrent+Depth+Approach10. Interpreting GPT: the logit lens (blog post) and Eliciting Latent Predictions from Transformers with the Tuned Lens — nostalgebraist; Nora Belrose et al., 2020 / 2023https://scholar.google.com/scholar?q=Interpreting+GPT%3A+the+logit+lens+%28blog+post%29+and+Eliciting+Latent+Predictions+from+Transformers+with+the+Tuned+Lens11. Do Language Models Use Their Depth Efficiently? — Róbert Csordás, Christopher D. Manning, Christopher Potts, 2025https://scholar.google.com/scholar?q=Do+Language+Models+Use+Their+Depth+Efficiently%3F12. The Curse of Depth in Large Language Models — Wenfang Sun, Xinyuan Song, Pengxiang Li, Lu Yin, Yefeng Zheng, Shiwei Liu, 2025https://scholar.google.com/scholar?q=The+Curse+of+Depth+in+Large+Language+Models13. Massive Activations in Large Language Models — Mingjie Sun, Xinlei Chen, J. Zico Kolter, Zhuang Liu, 2024https://scholar.google.com/scholar?q=Massive+Activations+in+Large+Language+Models14. Highway and Residual Networks Learn Unrolled Iterative Estimation — Klaus Greff, Rupesh K. Srivastava, Jürgen Schmidhuber, 2016https://scholar.google.com/scholar?q=Highway+and+Residual+Networks+Learn+Unrolled+Iterative+Estimation15. Draft & Verify: Lossless Large Language Model Acceleration via Self-Speculative Decoding — Jun Zhang et al., 2023https://scholar.google.com/scholar?q=Draft+%26+Verify%3A+Lossless+Large+Language+Model+Acceleration+via+Self-Speculative+Decoding16. Universal Transformers / Deep Equilibrium Models — Mostafa Dehghani et al.; Shaojie Bai, J. Zico Kolter, Vladlen Koltun, 2019https://scholar.google.com/scholar?q=Universal+Transformers+%2F+Deep+Equilibrium+Models17. Your Transformer is Secretly Linear — Anton Razzhigaev et al., 2024https://scholar.google.com/scholar?q=Your+Transformer+is+Secretly+Linear18. Layer by Layer: Uncovering Hidden Representations in Language Models — Oscar Skean et al., 2025https://scholar.google.com/scholar?q=Layer+by+Layer%3A+Uncovering+Hidden+Representations+in+Language+Models19. Transformer Feed-Forward Layers Are Key-Value Memories — Mor Geva, Roei Schuster, Jonathan Berant, Omer Levy, 2021https://scholar.google.com/scholar?q=Transformer+Feed-Forward+Layers+Are+Key-Value+Memories20. BERT Rediscovers the Classical NLP Pipeline — Ian Tenney, Dipanjan Das, Ellie Pavlick, 2019https://scholar.google.com/scholar?q=BERT+Rediscovers+the+Classical+NLP+Pipeline15 min
- Redwood: An AI-Designed Accelerator Built and Verified in Two WeeksThis episode examines "Redwood," a company technical report claiming an AI system took a two-architect spec to verified RTL for a spatial dataflow inference accelerator in under two weeks, with 95% coverage per block and a Qwen3-0.6B bring-up in week three. The hosts argue over whether the 14% first-silicon success statistic actually motivates the paper's single-spec approach, or whether handoffs between architecture, RTL, verification and kernels are the real schedule bottleneck. They explain why batch-size-one physical-AI inference is memory-bound rather than compute-bound, and walk through the tile design: RISC-V control core, matrix engine, vector engine, 512 KB scratchpad and a credit-based network-on-chip. Throughout, they separate the FPGA results measured on a Versal VPK180 from the Samsung 8 nm figures, which are projections. Those projections include the headline 1.75x throughput, 1.9x lower power and 3.4x performance-per-watt versus a Jetson Orin Nano, and the authors' claim of early recursive self-improvement. It's useful for listeners who want to judge AI-driven chip design claims by what was actually demonstrated.Sources:1. Redwood: A Frontier AI Accelerator Designed, Verified, and Deployed from Scratch in 2 Weeks by AI — Architect Labs, 2026http://arxiv.org/abs/2608.264182. ChipNeMo: Domain-Adapted LLMs for Chip Design — Mingjie Liu et al. (NVIDIA), 2023https://scholar.google.com/scholar?q=ChipNeMo%3A+Domain-Adapted+LLMs+for+Chip+Design3. Pushing the Limits of Machine Design: Automated CPU Design with AI — Shuyao Cheng et al., 2023https://scholar.google.com/scholar?q=Pushing+the+Limits+of+Machine+Design%3A+Automated+CPU+Design+with+AI4. Chip-Chat: Challenges and Opportunities in Conversational Hardware Design — Jason Blocklove, Siddharth Garg, Ramesh Karri, Hammond Pearce, 2023https://scholar.google.com/scholar?q=Chip-Chat%3A+Challenges+and+Opportunities+in+Conversational+Hardware+Design5. VerilogEval: Evaluating Large Language Models for Verilog Code Generation — Mingjie Liu, Nathaniel Pinckney, Brucek Khailany, Haoxing Ren, 2023https://scholar.google.com/scholar?q=VerilogEval%3A+Evaluating+Large+Language+Models+for+Verilog+Code+Generation6. CVDP: Verilog Design Problems benchmark for RTL design and verification agents — Nathaniel Pinckney et al. (NVIDIA), 2025https://scholar.google.com/scholar?q=CVDP%3A+Verilog+Design+Problems+benchmark+for+RTL+design+and+verification+agents7. FVEval: Understanding Language Model Capabilities in Formal Verification of Digital Hardware — Minwoo Kang et al., 2024https://scholar.google.com/scholar?q=FVEval%3A+Understanding+Language+Model+Capabilities+in+Formal+Verification+of+Digital+Hardware8. Roofline: An Insightful Visual Performance Model for Multicore Architectures — Samuel Williams, Andrew Waterman, David Patterson, 2009https://scholar.google.com/scholar?q=Roofline%3A+An+Insightful+Visual+Performance+Model+for+Multicore+Architectures9. FlightLLM: Efficient Large Language Model Inference with a Complete Mapping Flow on FPGAs — Shulin Zeng et al., 2024https://scholar.google.com/scholar?q=FlightLLM%3A+Efficient+Large+Language+Model+Inference+with+a+Complete+Mapping+Flow+on+FPGAs10. MLPerf Power: Benchmarking the Energy Efficiency of Machine Learning Systems from Microwatts to Megawatts for Sustainable AI — Arya Tschand et al., 2024https://scholar.google.com/scholar?q=MLPerf+Power%3A+Benchmarking+the+Energy+Efficiency+of+Machine+Learning+Systems+from+Microwatts+to+Megawatts+for+Sustainable+AI11. An Empirical Study of Qwen3 Quantization — Xingyu Zheng et al., 2025https://scholar.google.com/scholar?q=An+Empirical+Study+of+Qwen3+Quantization12. SmoothQuant: Accurate and Efficient Post-Training Quantization for Large Language Models — Guangxuan Xiao et al., 2023https://scholar.google.com/scholar?q=SmoothQuant%3A+Accurate+and+Efficient+Post-Training+Quantization+for+Large+Language+Models13. AlphaEvolve: A coding agent for scientific and algorithmic discovery — Google DeepMind (Novikov et al.), 2025https://scholar.google.com/scholar?q=AlphaEvolve%3A+A+coding+agent+for+scientific+and+algorithmic+discovery14. KernelBench: Can LLMs Write Efficient GPU Kernels? — Anne Ouyang et al., 2025https://scholar.google.com/scholar?q=KernelBench%3A+Can+LLMs+Write+Efficient+GPU+Kernels%3F15. A graph placement methodology for fast chip design (and the follow-up critique 'Reevaluating Google's Reinforcement Learning for IC Macro Placement') — Azalia Mirhoseini et al. (2021); Igor Markov (2024), 2021/2024https://scholar.google.com/scholar?q=A+graph+placement+methodology+for+fast+chip+design+%28and+the+follow-up+critique+%27Reevaluating+Google%27s+Reinforcement+Learning+for+IC+Macro+Placement%27%2916. Gemmini: Enabling Systematic Deep-Learning Architecture Evaluation via Full-Stack Integration — Hasan Genc et al., 2021https://scholar.google.com/scholar?q=Gemmini%3A+Enabling+Systematic+Deep-Learning+Architecture+Evaluation+via+Full-Stack+Integration17 min
- Full-Bandwidth Transformer: Feeding Hidden States Back Into the StackThis episode examines "Full-bandwidth transformer," which asks whether a model can feed its entire top-layer hidden state, rather than just the roughly 17-bit sampled token, back into the bottom of the stack at every decoding step. It explains why the KV cache doesn't already solve this: attention is full-bandwidth horizontally, but a state at layer l can only be read by higher layers, and the top layer's output is never cached. The discussion also weighs the design against RNNs and the Feedback Transformer. Nothing is overwritten and the full KV cache is retained, but sequential dependence is the real tension, and it is compared with Coconut, which replaces tokens rather than augmenting them. The hosts then cover how the paper keeps training parallel with multi-pass, Jacobi-style training and a gated fusion of state and token embedding. They cover the pass-mix schedule that keeps decoding stable, where a 75/25 one- and two-pass mix diverges but adding 3% three-pass batches yields a plateau. They set the headline claims aside for scrutiny: about 1.5x effective tokens, matching baselines trained on 2x data, negligible decoding cost, and shorter reasoning traces. Listeners interested in scaling limits, chain-of-thought's depth bottleneck, and getting more out of each training token will find it useful.Sources:1. Full-bandwidth transformer — Xi Wang, Ziyang Cai, Zheng Zhan, Harry Dong, Ying Fan, Gustavo de Rosa, Tim Pearce, John Langford, 2026http://arxiv.org/abs/2608.088882. Addressing Some Limitations of Transformers with Feedback Memory — Angela Fan, Thibaut Lavril, Edouard Grave, Armand Joulin, Sainbayar Sukhbaatar, 2020 (arXiv; later versions 2021)https://scholar.google.com/scholar?q=Addressing+Some+Limitations+of+Transformers+with+Feedback+Memory3. Training Large Language Models to Reason in a Continuous Latent Space (Coconut) — Shibo Hao, Sainbayar Sukhbaatar, DiJia Su, Xian Li, Zhiting Hu, Jason Weston, Yuandong Tian, 2024https://scholar.google.com/scholar?q=Training+Large+Language+Models+to+Reason+in+a+Continuous+Latent+Space+%28Coconut%294. Scaling up Test-Time Compute with Latent Reasoning: A Recurrent Depth Approach — Jonas Geiping, Sean McLeish, Neel Jain, John Kirchenbauer, Siddharth Singh, Brian Bartoldson, Bhavya Kailkhura, Abhinav Bhatele, Tom Goldstein, 2025https://scholar.google.com/scholar?q=Scaling+up+Test-Time+Compute+with+Latent+Reasoning%3A+A+Recurrent+Depth+Approach5. Chain of Thought Empowers Transformers to Solve Inherently Serial Problems — Zhiyuan Li, Hong Liu, Denny Zhou, Tengyu Ma, 2024https://scholar.google.com/scholar?q=Chain+of+Thought+Empowers+Transformers+to+Solve+Inherently+Serial+Problems6. Think before you speak: Training Language Models With Pause Tokens — Sachin Goyal, Ziwei Ji, Ankit Singh Rawat, Aditya Krishna Menon, Sanjiv Kumar, Vaishnavh Nagarajan, 2023https://scholar.google.com/scholar?q=Think+before+you+speak%3A+Training+Language+Models+With+Pause+Tokens7. PonderLM-2: Pretraining LLM with Latent Thoughts in Continuous Space — Boyi Zeng et al., 2025https://scholar.google.com/scholar?q=PonderLM-2%3A+Pretraining+LLM+with+Latent+Thoughts+in+Continuous+Space8. ParaRNN: Unlocking Parallel Training of Nonlinear RNNs for Large Language Models — Federico Danieli et al., 2025https://scholar.google.com/scholar?q=ParaRNN%3A+Unlocking+Parallel+Training+of+Nonlinear+RNNs+for+Large+Language+Models9. Parallelizing Non-linear Sequential Models over the Sequence Length (DEER) — Yi Heng Lim et al., 2024https://scholar.google.com/scholar?q=Parallelizing+Non-linear+Sequential+Models+over+the+Sequence+Length+%28DEER%2910. Accelerating Feedforward Computation via Parallel Nonlinear Equation Solving — Yang Song et al., 2021https://scholar.google.com/scholar?q=Accelerating+Feedforward+Computation+via+Parallel+Nonlinear+Equation+Solving11. The Expressive Power of Transformers with Chain of Thought — William Merrill, Ashish Sabharwal, 2024https://scholar.google.com/scholar?q=The+Expressive+Power+of+Transformers+with+Chain+of+Thought12. Let's Think Dot by Dot: Hidden Computation in Transformer Language Models — Jacob Pfau, William Merrill, Samuel R. Bowman, 2024https://scholar.google.com/scholar?q=Let%27s+Think+Dot+by+Dot%3A+Hidden+Computation+in+Transformer+Language+Models13. Reasoning with Latent Thoughts: On the Power of Looped Transformers — Nikunj Saunshi et al., 2025https://scholar.google.com/scholar?q=Reasoning+with+Latent+Thoughts%3A+On+the+Power+of+Looped+Transformers14. Scaling Latent Reasoning via Looped Language Models (Ouro) — Rui-Jie Zhu et al., 2025https://scholar.google.com/scholar?q=Scaling+Latent+Reasoning+via+Looped+Language+Models+%28Ouro%2915. Scaling Data-Constrained Language Models — Niklas Muennighoff et al., 2023https://scholar.google.com/scholar?q=Scaling+Data-Constrained+Language+Models16. Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety — Tomek Korbak et al., 2025https://scholar.google.com/scholar?q=Chain+of+Thought+Monitorability%3A+A+New+and+Fragile+Opportunity+for+AI+Safety17. Language Models Don't Always Say What They Think: Unfaithful Explanations in Chain-of-Thought Prompting — Miles Turpin et al., 2023https://scholar.google.com/scholar?q=Language+Models+Don%27t+Always+Say+What+They+Think%3A+Unfaithful+Explanations+in+Chain-of-Thought+Prompting18. The Illusion of State in State-Space Models — William Merrill, Jackson Petty, Ashish Sabharwal, 2024https://scholar.google.com/scholar?q=The+Illusion+of+State+in+State-Space+Models19. Block-Recurrent Transformers — DeLesley Hutchins et al., 2022https://scholar.google.com/scholar?q=Block-Recurrent+TransformersInteractive Visualization: Full-Bandwidth Transformer: Feeding Hidden States Back Into the Stack17 min
- The Matthew Effect in RL: Rich Get Richer on Hard ProblemsThis episode examines "Learning to Solve Hard Problems in RL for LLMs by Never Giving Up," which finds that RL post-training on math, code, and agentic coding benchmarks disproportionately improves performance on problems models already handle reasonably well, while barely moving the needle on the hardest tasks — a pattern the authors dub the Matthew Effect, after Robert Merton's sociology of science concept. The discussion contrasts two explanations for this skew: the intuitive "signal loss" account, where GRPO's group-relative reward gives zero gradient when every sampled completion fails a hard problem, versus the paper's "signal efficiency" argument, that compute is instead wasted reconfirming easy problems the model has already solved. That distinction matters because it points to different fixes — simply sampling more completions per prompt doesn't help, but reallocating sampling toward unsolved problems does, which motivates the paper's proposed method of persistently re-sampling unsolved prompts rather than discarding them. Listeners get a concrete walkthrough of the controlled GSM8k experiment used to test these competing hypotheses, including a difficulty-tiered evaluation and a K-sweep that challenges conventional assumptions about RL sample scaling. The episode is a useful listen for anyone weighing whether reinforcement learning can actually push language models past their pretrained capability ceiling, or whether it's mainly sharpening skills the model already has.Sources:1. Learning to Solve Hard Problems in RL for LLMs by Never Giving Up — Michael Noukhovitch, Hamish Ivison, Nathan Lambert, Aaron Courville, 2026http://arxiv.org/abs/2609.13443v12. Prioritized Experience Replay — Tom Schaul, John Quan, Ioannis Antonoglou, David Silver, 2015https://scholar.google.com/scholar?q=Prioritized+Experience+Replay3. DAPO: An Open-Source LLM Reinforcement Learning System at Scale — Qiying Yu, Zheng Zhang, et al. (ByteDance Seed), 2025https://scholar.google.com/scholar?q=DAPO%3A+An+Open-Source+LLM+Reinforcement+Learning+System+at+Scale4. Prioritized Level Replay — Minqi Jiang, Edward Grefenstette, Tim Rocktäschel, 2021https://scholar.google.com/scholar?q=Prioritized+Level+Replay5. Curriculum Learning — Yoshua Bengio, Jérôme Louradour, Ronan Collobert, Jason Weston, 2009https://scholar.google.com/scholar?q=Curriculum+Learning6. Never Give Up: Learning Directed Exploration Strategies — A. P. Badia, P. Sprechmann, A. Vitvitskyi, D. Guo, B. Piot, S. Kapturowski, O. Tieleman, M. Arjovsky, A. Pritzel, A. Bolt, C. Blundell, 2020https://scholar.google.com/scholar?q=Never+Give+Up%3A+Learning+Directed+Exploration+Strategies7. Reinforce-Ada: An Adaptive Sampling Framework under Non-linear RL Objectives — W. Xiong, C. Ye, B. Liao, H. Dong, X. Xu, C. Monz, J. Bian, N. Jiang, T. Zhang, 2025https://scholar.google.com/scholar?q=Reinforce-Ada%3A+An+Adaptive+Sampling+Framework+under+Non-linear+RL+Objectives8. POLARIS: A Post-Training Recipe for Scaling Reinforcement Learning on Advanced Reasoning Models — C. An, Z. Xie, X. Li, L. Li, J. Zhang, S. Gong, M. Zhong, J. Xu, X. Qiu, M. Wang, L. Kong, 2025https://scholar.google.com/scholar?q=POLARIS%3A+A+Post-Training+Recipe+for+Scaling+Reinforcement+Learning+on+Advanced+Reasoning+Models9. RL Grokking Recipe: How Does RL Unlock and Transfer New Algorithms in LLMs? — Y. Sun, Y. Cao, P. Huang, H. Bai, H. Hajishirzi, N. Dziri, D. Song, 2025https://scholar.google.com/scholar?q=RL+Grokking+Recipe%3A+How+Does+RL+Unlock+and+Transfer+New+Algorithms+in+LLMs%3F10. The Art of Scaling Reinforcement Learning Compute for LLMs — D. Khatri, L. Madaan, R. Tiwari, R. Bansal, S. S. Duvvuri, M. Zaheer, I. S. Dhillon, D. Brandfonbrener, R. Agarwal, 2025https://scholar.google.com/scholar?q=The+Art+of+Scaling+Reinforcement+Learning+Compute+for+LLMs11. The Primacy Bias in Deep Reinforcement Learning — E. Nikishin, M. Schwarzer, P. D'Oro, P.-L. Bacon, A. Courville, 2022https://scholar.google.com/scholar?q=The+Primacy+Bias+in+Deep+Reinforcement+Learning12. Act Only When It Pays: Efficient Reinforcement Learning for LLM Reasoning via Selective Rollouts (GRESO) — H. Zheng, Y. Zhou, B. R. Bartoldson, B. Kailkhura, F. Lai, J. Zhao, B. Chen, 2025https://scholar.google.com/scholar?q=Act+Only+When+It+Pays%3A+Efficient+Reinforcement+Learning+for+LLM+Reasoning+via+Selective+Rollouts+%28GRESO%29Interactive Visualization: The Matthew Effect in RL: Rich Get Richer on Hard Problems15 min
- NVQLink: Wiring Supercomputers Directly to Quantum ChipsThis episode examines the NVQLink architecture, a NVIDIA-led collaboration with nine national labs and research institutions that tightly couples high-performance computing with quantum processors. The discussion centers on why reaction time—not just throughput—is critical for quantum error correction, since qubits decohere while waiting for classical correction signals to arrive. It breaks down the system's building blocks (QPU, QSC, PPU, and the real-time host) and explores a counterintuitive design choice: routing the real-time interconnect over commodity, unreliable Ethernet rather than PCIe, trading a seemingly obvious direct connection for datacenter-scale networking. The episode also connects the work to a decade-old 2017 paper by co-author Travis Humble that first framed QPUs as HPC accelerators, showing how NVQLink turns that early concept into a measured, functioning system with sub-4-microsecond round-trip latency. Listeners interested in the intersection of classical computing infrastructure and quantum hardware will find the engineering tradeoffs and vocabulary breakdown a useful entry point into a fast-moving, unfamiliar corner of systems design.Sources:1. Platform Architecture for Tight Coupling of High-Performance Computing with Quantum Processors — Shane A. Caldwell, Moein Khazraee, Elena Agostini, Tom Lassiter, Corey Simpson, Omri Kahalon, Mrudula Kanuri, Jin-Sung Kim, Sam Stanwyck, Muyuan Li, Jan Olle, Christopher Chamberland, Ben Howe, Bruno Schmitt, Justin G. Lietz, Alex McCaskey, Jun Ye, Ang Li, Alicia B. Magann, Corey I. Ostrove, Kenneth Rudinger, Robin Blume-Kohout, Kevin Young, Nathan E. Miller, Yilun Xu, Gang Huang, Irfan Siddiqi, John Lange, Christopher Zimmer, Travis Humble, 2025http://arxiv.org/abs/2510.252132. High-Performance Computing with Quantum Processing Units — K. Britt, T. Humble, 2017https://scholar.google.com/scholar?q=High-Performance+Computing+with+Quantum+Processing+Units3. CUDA Quantum: The Platform for Integrated Quantum-Classical Computing — A. McCaskey et al. (NVIDIA), 2024https://scholar.google.com/scholar?q=CUDA+Quantum%3A+The+Platform+for+Integrated+Quantum-Classical+Computing4. Quantum-Centric Supercomputing for Materials Science: A Perspective on Challenges and Future Directions — Y. Alexeev et al. (IBM, national labs), 2021 (Future Generation Computer Systems)https://scholar.google.com/scholar?q=Quantum-Centric+Supercomputing+for+Materials+Science%3A+A+Perspective+on+Challenges+and+Future+Directions5. MLIR: Scaling Compiler Infrastructure for Domain Specific Computation — C. Lattner, M. Amini, U. Bondhugula, A. Cohen, A. Davis, J. Pienaar, R. Riddle, T. Shpeisman, N. Vasilache, O. Zinenko (Google), 2021 (CGO)https://scholar.google.com/scholar?q=MLIR%3A+Scaling+Compiler+Infrastructure+for+Domain+Specific+Computation6. QICK: Quantum Instrumentation Control Kit — L. Stefanazzi, K. Treptow, N. Wilcer, et al. (Fermilab), 2022 (Review of Scientific Instruments)https://scholar.google.com/scholar?q=QICK%3A+Quantum+Instrumentation+Control+Kit7. Real-Time Processing Systems for Next-Generation Quantum Computers (representative of the QEC-decoder-latency literature, e.g. Google Quantum AI's real-time decoding work) — Google Quantum AI et al., 2023–2024https://scholar.google.com/scholar?q=Real-Time+Processing+Systems+for+Next-Generation+Quantum+Computers+%28representative+of+the+QEC-decoder-latency+literature%2C+e.g.+Google+Quantum+AI%27s+real-time+decoding+work%298. Learning high-accuracy error decoding for quantum processors (AlphaQubit) — Johannes Bausch, Andrew W. Senior, Francisco J. H. Heras, et al., 2024https://scholar.google.com/scholar?q=Learning+high-accuracy+error+decoding+for+quantum+processors+%28AlphaQubit%299. Evaluation of the Classical Hardware Requirements for Large-Scale Quantum Computations — Daan Camps, Ermal Rrapaj, Katherine Klymko, Brian Austin, Nicholas J. Wright, 2024https://scholar.google.com/scholar?q=Evaluation+of+the+Classical+Hardware+Requirements+for+Large-Scale+Quantum+Computations10. Parallel window decoding enables scalable fault tolerant quantum computation — Luka Skoric, Dan E. Browne, Kenton M. Barnes, Neil I. Gillespie, Earl T. Campbell, 2023https://scholar.google.com/scholar?q=Parallel+window+decoding+enables+scalable+fault+tolerant+quantum+computation11. Quantum error correction below the surface code threshold — Google Quantum AI and Collaborators, 2024https://scholar.google.com/scholar?q=Quantum+error+correction+below+the+surface+code+threshold12. A Game of Surface Codes: Large-Scale Quantum Computing with Lattice Surgery — Daniel Litinski, 2019https://scholar.google.com/scholar?q=A+Game+of+Surface+Codes%3A+Large-Scale+Quantum+Computing+with+Lattice+SurgeryInteractive Visualization: NVQLink: Wiring Supercomputers Directly to Quantum Chips19 min
- ProgramBench: Frontier Agents Fail to Rebuild Software from ScratchThis episode examines ProgramBench, a new benchmark testing whether frontier language models can rebuild working software from just a compiled binary and its documentation, with no source code, scaffolding, or prescribed architecture to work from. Across nine frontier models and two hundred tasks spanning CLI tools up to FFmpeg, SQLite, and the PHP interpreter, zero tasks were fully resolved, exposing a stark gap between patching existing code and making the upstream architectural decisions—language choice, module boundaries, data structures, error handling—that real software design requires. The discussion unpacks the benchmark's clever self-hosting trick: an LLM agent fuzzes the reference binary to build a behavioral test suite, enabling black-box grading that judges programs by what they do rather than how closely they mimic the original source. Framing the results against Parnas's classic work on information hiding and modular decomposition, the conversation argues that current agents default to monolithic, unstructured code once nobody hands them a skeleton to fill in. It's a sobering data point for anyone assuming coding agents are close to functioning as autonomous software architects rather than sophisticated patch-writers.Sources:1. ProgramBench: Can Language Models Rebuild Programs From Scratch? — John Yang, Kilian Lieret, Jeffrey Ma, Parth Thakkar, Dmitrii Pedchenko, Sten Sootla, Emily McMilin, Pengcheng Yin, Rui Hou, Gabriel Synnaeve, Diyi Yang, Ofir Press, 2026http://arxiv.org/abs/2605.035462. On the Criteria To Be Used in Decomposing Systems into Modules — D. L. Parnas, 1972https://scholar.google.com/scholar?q=On+the+Criteria+To+Be+Used+in+Decomposing+Systems+into+Modules3. Measuring Coding Challenge Competence With APPS — Dan Hendrycks, Steven Basart, Saurav Kadavath, et al., 2021https://scholar.google.com/scholar?q=Measuring+Coding+Challenge+Competence+With+APPS4. SWE-bench: Can Language Models Resolve Real-World GitHub Issues? — Carlos E. Jimenez, John Yang, Alexander Wettig, et al., 2024https://scholar.google.com/scholar?q=SWE-bench%3A+Can+Language+Models+Resolve+Real-World+GitHub+Issues%3F5. SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering — John Yang, Carlos E. Jimenez, Alexander Wettig, et al., 2024https://scholar.google.com/scholar?q=SWE-agent%3A+Agent-Computer+Interfaces+Enable+Automated+Software+Engineering6. QuickCheck: A Lightweight Tool for Random Testing of Haskell Programs — Koen Claessen, John Hughes, 2000https://scholar.google.com/scholar?q=QuickCheck%3A+A+Lightweight+Tool+for+Random+Testing+of+Haskell+Programs7. Finding and Understanding Bugs in C Compilers — Xuejun Yang, Yang Chen, Eric Eide, John Regehr, 2011https://scholar.google.com/scholar?q=Finding+and+Understanding+Bugs+in+C+Compilers8. CodeT: Code Generation with Generated Tests — Bei Chen, Fengji Zhang, Anh Nguyen, et al., 2022https://scholar.google.com/scholar?q=CodeT%3A+Code+Generation+with+Generated+Tests9. Commit0: Library Generation from Scratch — Wenting Zhao, Nan Jiang, Celine Lee, Justin T Chiu, Claire Cardie, Matthias Gallé, Alexander M Rush, 2024https://scholar.google.com/scholar?q=Commit0%3A+Library+Generation+from+Scratch10. DevBench: A Comprehensive Benchmark for Software Development — Bowen Li, Wenhan Wu, Ziwei Tang, et al. (incl. John Yang, Ofir Press), 2024https://scholar.google.com/scholar?q=DevBench%3A+A+Comprehensive+Benchmark+for+Software+Development11. NL2Repo-bench: Towards Long-Horizon Repository Generation Evaluation of Coding Agents — Jingzhe Ding et al., 2026https://scholar.google.com/scholar?q=NL2Repo-bench%3A+Towards+Long-Horizon+Repository+Generation+Evaluation+of+Coding+Agents12. LLM4Decompile: Decompiling Binary Code with Large Language Models — Hanzhuo Tan, Qi Luo, Jing Li, Yuqun Zhang, 2024https://scholar.google.com/scholar?q=LLM4Decompile%3A+Decompiling+Binary+Code+with+Large+Language+Models13. Position: Humans Are Missing from AI Coding Agent Research — Zora Zhiruo Wang, John Yang, Kilian Lieret, et al., 2026https://scholar.google.com/scholar?q=Position%3A+Humans+Are+Missing+from+AI+Coding+Agent+ResearchInteractive Visualization: ProgramBench: Frontier Agents Fail to Rebuild Software from Scratch19 min
- Recursive Self-Improvement and Alignment: OpenAI's "An Alien Mind"This episode examines "An Alien Mind," a single-author essay from an OpenAI research leader arguing that internal results point toward sustained progress and eventually recursive self-improvement, and asks what an outside reader could actually verify. The hosts note that the essay offers no methods, tables, or error bars. They contrast its scaling claims with quantitative work like the Kaplan and Hoffmann scaling-law papers, which give fitted curves and exponents. They also discuss the essay's admission that easy-to-measure capabilities improve faster than hard-to-quantify ones, which makes progress harder to gauge. A large part of the discussion covers how the essay defines alignment: goal alignment versus value alignment, and whether that split is a real testable distinction or a blurry one. The hosts also introduce chain-of-thought monitoring, its fragility when reasoning is trained to look good, and safety cases as a basis for mandated bars. Listeners get a skeptical, evidence-focused look at what a lab insider's claims about AI progress and safety would need in order to hold up.Sources:1. Recursive Self-Improvement and Alignment: OpenAI's "An Alien Mind"https://openai.com/index/an-alien-mind/2. Artificial Intelligence, Values, and Alignment — Iason Gabriel, 2020https://scholar.google.com/scholar?q=Artificial+Intelligence%2C+Values%2C+and+Alignment3. Training language models to follow instructions with human feedback (InstructGPT) — Long Ouyang, Jeff Wu, Xu Jiang, Diogo Almeida, Carroll Wainwright, Pamela Mishkin, et al. (OpenAI), 2022https://scholar.google.com/scholar?q=Training+language+models+to+follow+instructions+with+human+feedback+%28InstructGPT%294. The Instruction Hierarchy: Training LLMs to Prioritize Privileged Instructions — Eric Wallace, Kai Xiao, Reimar Leike, Lilian Weng, Johannes Heidecke, Alex Beutel (OpenAI), 2024https://scholar.google.com/scholar?q=The+Instruction+Hierarchy%3A+Training+LLMs+to+Prioritize+Privileged+Instructions5. Emergent Misalignment: Narrow Finetuning Can Produce Broadly Misaligned LLMs — Jan Betley, Daniel Tan, Niels Warncke, Anna Sztyber-Betley, Xuchan Bao, Martín Soto, Nathan Labenz, Owain Evans, 2025https://scholar.google.com/scholar?q=Emergent+Misalignment%3A+Narrow+Finetuning+Can+Produce+Broadly+Misaligned+LLMs6. Chain of Thought Monitorability: A New and Fragile Opportunity for AI Safety — Korbak et al. (multi-lab position paper), 2025https://scholar.google.com/scholar?q=Chain+of+Thought+Monitorability%3A+A+New+and+Fragile+Opportunity+for+AI+Safety7. Monitoring Reasoning Models for Misbehavior and the Risks of Promoting Obfuscation — Baker et al. (OpenAI), 2025https://scholar.google.com/scholar?q=Monitoring+Reasoning+Models+for+Misbehavior+and+the+Risks+of+Promoting+Obfuscation8. Clarifying AI alignment — Paul Christiano, 2018https://scholar.google.com/scholar?q=Clarifying+AI+alignment9. Deep Reinforcement Learning from Human Preferences — Christiano, Leike, Brown, Martic, Legg, Amodei, 2017https://scholar.google.com/scholar?q=Deep+Reinforcement+Learning+from+Human+Preferences10. Alignment Faking in Large Language Models — Greenblatt et al. (Anthropic and Redwood Research), 2024https://scholar.google.com/scholar?q=Alignment+Faking+in+Large+Language+Models11. The Persona Selection Model — Anthropic, 2026https://scholar.google.com/scholar?q=The+Persona+Selection+Model12. Goal Misgeneralization in Deep Reinforcement Learning — Langosco, Koch, Sharkey, Pfau, Krueger, 2022https://scholar.google.com/scholar?q=Goal+Misgeneralization+in+Deep+Reinforcement+Learning13. Risks from Learned Optimization in Advanced Machine Learning Systems — Hubinger, van Merwe, Mikulik, Skalse, Garrabrant, 2019https://scholar.google.com/scholar?q=Risks+from+Learned+Optimization+in+Advanced+Machine+Learning+Systems14. Weak-to-Strong Generalization: Eliciting Strong Capabilities with Weak Supervision — Burns et al. (OpenAI), 2023https://scholar.google.com/scholar?q=Weak-to-Strong+Generalization%3A+Eliciting+Strong+Capabilities+with+Weak+Supervision15. Training Language Models to Self-Report / Confessions — OpenAI Alignment team, 2025-2026https://scholar.google.com/scholar?q=Training+Language+Models+to+Self-Report+%2F+ConfessionsInteractive Visualization: Recursive Self-Improvement and Alignment: OpenAI's "An Alien Mind"16 min
- Latent Space Communication via K-V Cache AlignmentThis episode explores a Google DeepMind paper proposing that separately trained language models can exchange raw internal state — the key-value cache built during transformer inference — through a shared "global latent space," rather than communicating only through text. The hosts unpack why text is a lossy bottleneck for inter-model communication, and how lightweight, frozen-weight adapter pairs let each model translate its own cache into and out of this common space, keeping training cost linear rather than combinatorial as more models join the pool. A striking result anchors the discussion: translating a model's cache through this shared space can sometimes outperform the model's own untouched cache on the same task. The conversation connects this idea to familiar concepts like prefix-tuning and continuous latent reasoning, framing the cache exchange as a dynamic, evolving version of a static soft prompt. Listeners interested in how models might one day share "trains of thought" instead of finished sentences will find the tension between the approach's architectural simplicity and its surprising performance gains especially compelling.Sources:1. Latent Space Communication via K-V Cache Alignmenthttps://arxiv.org/pdf/2601.061232. Prefix-Tuning: Optimizing Continuous Prompts for Generation — Xiang Lisa Li, Percy Liang, 2021https://scholar.google.com/scholar?q=Prefix-Tuning%3A+Optimizing+Continuous+Prompts+for+Generation3. The Power of Scale for Parameter-Efficient Prompt Tuning — Brian Lester, Rami Al-Rfou, Noah Constant, 2021https://scholar.google.com/scholar?q=The+Power+of+Scale+for+Parameter-Efficient+Prompt+Tuning4. Relative Representations Enable Zero-Shot Latent Space Communication — Luca Moschella, Valentino Maiorca, Marco Fumero, Antonio Norelli, Francesco Locatello, Emanuele Rodolà, 2022https://scholar.google.com/scholar?q=Relative+Representations+Enable+Zero-Shot+Latent+Space+Communication5. Git Re-Basin: Merging Models modulo Permutation Symmetries — Samuel K. Ainsworth, Jonathan Hayase, Siddhartha Srinivasa, 2022https://scholar.google.com/scholar?q=Git+Re-Basin%3A+Merging+Models+modulo+Permutation+Symmetries6. On the direct alignment of latent spaces — Lähner, Moeller, 2024https://scholar.google.com/scholar?q=On+the+direct+alignment+of+latent+spaces7. Harnessing the universal geometry of embeddings — Jha, Zhang, Shmatikov, Morris, 2025https://scholar.google.com/scholar?q=Harnessing+the+universal+geometry+of+embeddings8. Training Large Language Models to Reason in a Continuous Latent Space (Coconut) — Hao, Sukhbaatar, Su, Li, Hu, Weston, Tian, 2024https://scholar.google.com/scholar?q=Training+Large+Language+Models+to+Reason+in+a+Continuous+Latent+Space+%28Coconut%299. DiPaCo: Distributed Path Composition — Douillard, Feng, Rusu, Kuncoro, Donchev, Chhaparia, Gog, Ranzato, Shen, Szlam, 2024https://scholar.google.com/scholar?q=DiPaCo%3A+Distributed+Path+CompositionInteractive Visualization: Latent Space Communication via K-V Cache Alignment20 min
- Dream-RSI: Teaching AI How to Search, Not Just SolveThis episode examines "Dream-RSI: Recursive Self-Improvement through Evolving Worlds," which proposes making the exploration strategy of a discovery system — not the underlying model — the target of recursive self-improvement. Rather than optimizing candidate solutions directly, the system optimizes the policy that decides how to search: which branches to expand, how to parallelize workers, and when to stop, while the underlying coding agent (Gemini, in the paper's experiments) stays fixed. The key innovation is "dreaming": replaying an already-recorded discovery tree of past generate-evaluate attempts as a cheap simulator, letting new exploration policies be scored for free against historical outcomes instead of running costly new agent calls. This produces a three-stage loop — online exploration to grow the tree, constructing a replay simulator from it, then "dreaming" to test and select better policies before redeploying them — addressing the classic problem that policy-level exploration research suffers from painfully delayed feedback. The discussion situates the work against prior exploration methods like bandit algorithms, RL², Never Give Up, and FunSearch, making it a useful listen for anyone interested in how search-strategy meta-optimization, rather than raw model capability, might be the next lever for scaling AI-driven discovery.Sources:1. Dream-RSI: Recursive Self-Improvement through Evolving Worlds — Tong Zheng, Xidong Wu, Zheng Zhang, Zhankui He, Chaoyi Zhang, Benjamin Coleman, Ruoqiao Wei, Di Bai, Haolin Liu, Rui Liu, Xue Wang, Yue Zhuan, Wang-Cheng Kang, Renkai Xiang, Heng Huang, Xinwu Cheng, Yunsong Guo, 2026http://arxiv.org/abs/2609.148582. RL²: Fast Reinforcement Learning via Slow Reinforcement Learning — Yan Duan, John Schulman, Xi Chen, Peter L. Bartlett, Ilya Sutskever, Pieter Abbeel, 2016https://scholar.google.com/scholar?q=RL%C2%B2%3A+Fast+Reinforcement+Learning+via+Slow+Reinforcement+Learning3. Never Give Up: Learning Directed Exploration Strategies — Adrià Puigdomènech Badia, Pablo Sprechmann, Alex Vitvitskyi, et al. (DeepMind), 2020https://scholar.google.com/scholar?q=Never+Give+Up%3A+Learning+Directed+Exploration+Strategies4. Mathematical discoveries from program search with large language models (FunSearch) — Bernardino Romera-Paredes, Mohammadamin Barekatain, Alexander Novikov, et al. (Google DeepMind), 2024https://scholar.google.com/scholar?q=Mathematical+discoveries+from+program+search+with+large+language+models+%28FunSearch%295. Taking the Human Out of the Loop: A Review of Bayesian Optimization — Bobak Shahriari, Kevin Swersky, Ziyu Wang, Ryan P. Adams, Nando de Freitas, 2016https://scholar.google.com/scholar?q=Taking+the+Human+Out+of+the+Loop%3A+A+Review+of+Bayesian+Optimization6. AlphaEvolve: A coding agent for scientific and algorithmic discovery — A. Novikov et al., 2025https://scholar.google.com/scholar?q=AlphaEvolve%3A+A+coding+agent+for+scientific+and+algorithmic+discovery7. Mastering diverse domains through world models (Dreamer V3) — D. Hafner, J. Pasukonis, J. Ba, T. Lillicrap, 2023https://scholar.google.com/scholar?q=Mastering+diverse+domains+through+world+models+%28Dreamer+V3%298. EvoX: Meta-evolution for automated discovery — S. Liu et al., 2026https://scholar.google.com/scholar?q=EvoX%3A+Meta-evolution+for+automated+discovery9. Evaluation-driven scaling for scientific discovery (SimpleTES) — H. Ye et al., 2026https://scholar.google.com/scholar?q=Evaluation-driven+scaling+for+scientific+discovery+%28SimpleTES%29Interactive Visualization: Dream-RSI: Teaching AI How to Search, Not Just Solve17 min
- CacheBridge: Fixing Cross-Model KV Cache Transfer FailuresThis episode examines CacheBridge, a paper proposing targeted fixes to a training-free method for transferring KV caches between different transformer models in multi-model routing setups. It explains why caches can't simply be handed off — differing residual widths, GQA head counts, and RoPE position encoding make one model's cache unreadable to another — and how a prior affine-mapper approach (FULL-HEADMAPPING) could swing wildly from near-native accuracy on one model pair to catastrophic collapse on another, with no way to predict which. The discussion breaks down the paper's four diagnosed failure causes, spanning head-mixing, mismatched error metrics, layer-count cost scaling, and a GPU implementation bottleneck, then covers the three corresponding repairs: HEAD-LOCAL's narrower one-to-one head mapping, ATTN-REPAIR's attention-aware calibration reweighting, and FUSED-FIT's custom kernel for building the mapper efficiently. Listeners interested in LLM serving infrastructure will find it a concrete look at diagnosing and patching a deployed technique rather than proposing a new architecture from scratch.Sources:1. CacheBridge: Efficient Cross-Model KV Cache Transfer — Xingyu Qu, Siyuan Lu, Zhiyu Chen, Sheng Wang, Tao Lin, 2026http://arxiv.org/abs/2609.008912. Cross-model KV cache transfer in LLM families: A closed-form linear mapping for prefill reuse — Heo, T., Shafipour, R., Zhao, R., Golub, M., Kamani, M. M., Borkar, R., Chandran, M. T., Zardoshti, P., Rouhani, B. D., 2026https://scholar.google.com/scholar?q=Cross-model+KV+cache+transfer+in+LLM+families%3A+A+closed-form+linear+mapping+for+prefill+reuse3. Cache-to-cache: Direct semantic communication between large language models — Fu, T., Min, Z., Zhang, H., Yan, J., Dai, G., Ouyang, W., Wang, Y., 2026https://scholar.google.com/scholar?q=Cache-to-cache%3A+Direct+semantic+communication+between+large+language+models4. Mixture-of-translators: Translating KV caches across heterogeneous large language models — Lee, J.-w., Song, M., Oh, J., Han, S., Park, S., Jang, G., Lim, S., 2026https://scholar.google.com/scholar?q=Mixture-of-translators%3A+Translating+KV+caches+across+heterogeneous+large+language+models5. DroidSpeak: KV cache sharing across fine-tuned model variants — Liu, Y., Huang, Y., Yao, J., Feng, S., Gu, Z., Du, K., Li, H., Cheng, Y., Jiang, J., Lu, S., Musuvathi, M., Choukse, E., 2026https://scholar.google.com/scholar?q=DroidSpeak%3A+KV+cache+sharing+across+fine-tuned+model+variants6. ICaRus: Identical cache reuse for efficient multi model inference — Woo, S., Kil, J., Kim, H., Kim, M., Kim, J., Seo, A., Lee, S., Jo, M., Ryu, J., Park, B., Kwon, S. J., Lee, D., 2026https://scholar.google.com/scholar?q=ICaRus%3A+Identical+cache+reuse+for+efficient+multi+model+inference7. GQA: Training generalized multi-query transformer models from multi-head checkpoints — Ainslie, J., Lee-Thorp, J., de Jong, M., Zemlyanskiy, Y., Lebrón, F., Sanghai, S., 2023https://scholar.google.com/scholar?q=GQA%3A+Training+generalized+multi-query+transformer+models+from+multi-head+checkpointsInteractive Visualization: CacheBridge: Fixing Cross-Model KV Cache Transfer Failures19 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.