Machine Learning Street Talk

→ WHAT IT COVERS Alistair Pullen, CEO of Cosine AI, explains how his UK-based frontier lab secured government compute allocation on the ISAMBARD supercomputer to build Britain's first sovereign LLM, why nations cannot rely on US AI access, and how constrained resources force architectural and algorithmic innovation in model training. → KEY INSIGHTS - **Sovereign AI compute strategy:** Cosine AI received government-allocated compute on the ISAMBARD supercomputer cluster in Bristol through the...

→ WHAT IT COVERS The winning team from the ARC-AGI-3 benchmark competition explains how they built a coding-agent harness using frontier LLMs to solve novel video games without instructions. The episode covers action efficiency scoring, LLM game priors, brute-force limitations, requirements-based engineering with AI coding agents, and whether benchmark performance correlates with genuine intelligence.

→ WHAT IT COVERS Thomas Ahle, researcher at Normal Computing, discusses thermodynamic computing chips that use electrical noise as computation rather than eliminating it, AI-assisted chip design using Verilog simulators built by swarms of agents, formal verification challenges in hardware, and the broader risks of AI-generated code eroding human understanding across engineering teams.

→ WHAT IT COVERS John Jumper, Nobel Prize-winning lead of DeepMind's AlphaFold team, explains how the system predicts protein structures in minutes instead of years, what it actually solves versus what remains unsolved, and why hybrid domain-specific AI architectures outperform general-purpose approaches in scientific discovery. → KEY INSIGHTS - **AlphaFold's actual scope:** AlphaFold predicts one specific class of scientific measurement — protein 3D structure from amino acid sequence — with...

→ WHAT IT COVERS Brad Carson, former U.S. Congressman and Department of Defense official, examines how AI reshapes military targeting, autonomous weapons accountability, deepfake liability, and AI regulation. He argues against technological fatalism, advocates for mandatory frontier model testing, US-China diplomatic engagement on AI, and congressional oversight to prevent informal Silicon Valley capture of AI policy.

→ WHAT IT COVERS UC Berkeley Professor Michael I. Jordan argues that AI development requires collective economic thinking rather than anthropomorphized intelligence narratives. He critiques AGI terminology as distortionary PR, advocates for mechanism design and game theory as frameworks for building AI systems, and warns that alarmist rhetoric from prominent researchers is actively demoralizing the next generation of technologists.

→ WHAT IT COVERS Beth Barnes and David Rein from METR explain their Time Horizons benchmark, which measures AI capability using human task-completion time as a unified metric spanning GPT-2 through current frontier models. They cover evaluation methodology, agentic scaffolding, reward hacking in capable models, and why extrapolating benchmark trends to real-world economic impact requires significant caution across multiple dimensions.

→ WHAT IT COVERS Robert Lange from Sakana AI discusses Shinka Evolve, an open-source evolutionary framework that uses multiple LLMs in parallel to discover novel algorithms and scientific solutions. The system improves on AlphaEvolve's approach through model ensembling, UCB-based adaptive model selection, and crossover mutations, achieving state-of-the-art circle packing results in under 200 LLM evaluations.

→ WHAT IT COVERS Deep learning pioneer Jeremy Howard joins Machine Learning Street Talk to argue that vibe coding functions like a slot machine, creating an illusion of control while eroding genuine software engineering competence. He draws on ULMFiT's origins, transfer learning history, and his own Claude Code experiments to distinguish coding from software engineering, warning that organizations betting on AI productivity gains face measurable, documented risks.

→ WHAT IT COVERS Blaise Agüera y Arcas presents research showing evolution can produce complex programs without mutation through symbiogenesis. Using BFF (Brain Fuck Forth) simulations with 1,024 random tapes of 64 bytes, he demonstrates how replicators merge to create computational complexity, experiencing phase transitions similar to gelation that transform random noise into functional life.

→ WHAT IT COVERS Dr. Jeff Beck explores energy-based models, variational autoencoders, and the nature of agency in AI systems. The conversation covers geometric deep learning, Bayesian inference, self-supervised learning architectures like JEPA, continual learning challenges, and the future of autonomous AI systems capable of scientific discovery and experimental design.

→ WHAT IT COVERS Philosopher Mazviita Chirimuuta examines how scientific abstraction and idealization shape neuroscience and AI research. She challenges computational theories of mind, argues biological cognition cannot be separated from living tissue, and presents haptic realism as an alternative to spectator theories of knowledge that assume mathematical representations reveal underlying universal truths.

→ WHAT IT COVERS Philosopher Mazviita Chiramuta challenges neuroscience's computational metaphors for the brain, arguing scientists mistake elegant simplifications for literal truth. The episode examines how every era models the mind using contemporary technology—from hydraulic pumps to computers—and questions whether Karl Friston's free energy principle and AI's inevitability represent genuine understanding or another historical illusion.

→ WHAT IT COVERS Dr. Jeff Beck explains why scaling Bayesian inference with object-centered models represents the path to human-like AI, contrasting structured cognitive approaches with current transformer architectures that lack explicit world models and causal reasoning capabilities. → KEY INSIGHTS - **Bayesian Brain Evidence:** Humans perform optimal cue combination in sensory-motor tasks, adjusting for reliability on a trial-by-trial basis without knowing which sensory input is more...

→ WHAT IT COVERS Max Bennett explains how the brain evolved through five breakthroughs, from basic steering to mental simulation, revealing how the neocortex functions as a generative model that enables planning, imagination, and social cognition through 600 million years of evolution. → KEY INSIGHTS - **Perception as Inference:** The brain does not directly perceive sensory input but constructs models of reality and tests them against evidence.

→ WHAT IT COVERS César Hidalgo presents three laws governing knowledge growth, diffusion, and value, demonstrating how knowledge accumulates through experience following power laws, diffuses through geographic and social networks based on relatedness, and requires physical embodiment in teams and organizations rather than existing abstractly in documents.

→ WHAT IT COVERS Dr. Mike Israetel debates artificial superintelligence timelines, predicting ASI arrives in 2026-2027 before AGI in 2029-2031. Discussion covers intelligence definitions, embodied cognition versus abstraction, reasoning capabilities, live learning challenges, and whether current AI systems truly understand versus mimic. → KEY INSIGHTS - **ASI Timeline Prediction:** Israetel predicts artificial superintelligence emerges late 2026 when AI systems demonstrate 10x-100x human...

→ WHAT IT COVERS Category theory provides a mathematical framework for designing neural networks that can reliably execute algorithms like addition and multiplication, addressing fundamental limitations in current large language models and deep learning architectures. → KEY INSIGHTS - **Algorithmic Failure in LLMs:** Large language models perform hundreds of billions of multiplications to generate single tokens yet cannot reliably multiply small numbers together, revealing misalignment between...

→ WHAT IT COVERS Andrew Gordon and Nora Petrova from Prolific explain why current AI benchmarks miss critical user experience factors and introduce their human-centered evaluation methodology called Humane. → KEY INSIGHTS - **TrueSkill Methodology:** Prolific uses Microsoft's TrueSkill framework from Xbox Live to run AI model tournaments, selecting model pairs based on information gain to minimize uncertainty efficiently with fewer comparisons needed.

→ WHAT IT COVERS Professor Yi Ma presents a mathematical theory of intelligence based on parsimony and self-consistency principles, explaining how compression drives knowledge acquisition across evolutionary, neural, and scientific stages while deriving white-box transformer architectures from first principles. → KEY INSIGHTS - **Rate Reduction Framework:** Intelligence operates by discovering low-dimensional structures in high-dimensional data through compression, where the coding rate...