Machine Learning Street Talk

Tim Scarfe × Alistair Pullen

Cosine’s sovereign-AI strategy pairs UK-funded training compute with customer-owned inference, making a narrow, capital-disciplined build possible without financing token-serving infrastructure.Performance differentiation is shifting toward active parameters, post-training data and large-scale RL, but reward attribution, software verification, swarm complexity and export-control hardware dependence remain material execution risks.

Thomas Ahle × Tim Scarfe

Normal Computing’s CN101 uses noise-driven capacitor and programmable-resistance arrays to implement stochastic differential equations for a narrow class of probabilistic workloads.The commercial question is whether scaled benchmarks and co-designed models can convert component-level speedups into system-level gains, while AI-assisted EDA still faces a correctness and formalized-intent trust exercise.

Maxwell Ramstead × Guillaume Verdon

Extropic is building mixed-signal silicon that harnesses stochastic electron dynamics for probabilistic workloads, challenging deterministic hardware’s escalating energy cost of suppressing noise.Its reported 300-degree-of-freedom prototype and projections for 1,000–100,000X chip-level efficiency remain early, with heterogeneous adoption more likely than immediate GPU replacement.

Eiso Kant

Poolside is betting that reinforcement learning from execution feedback across nearly 1 million repositories can deliver human-level intelligence across most knowledge work within Kant’s 18–36-month estimate.Its moat is cumulative data, infrastructure, talent, and enterprise distribution, while current developer productivity gains remain closer to 20–30% and the shift from AI-assisted work to autonomous software teams remains uneven.

John Palazza

CentML targets stranded GPU capacity, where organizations often use only 30%-40% of a GPU versus a potential 80%-100%, while enterprise pilots become economically difficult at production scale.Its hardware- and cloud-agnostic platform claims savings of up to 60%, and rising efficiency could fund more AI workloads, agents, and open-weight deployments rather than reduce compute demand.