Tom McGrath

Erik Torenberg × Nathan Labenz × Geoffrey Irving × Daniel Murfet × prinz × Rahul Sonwalkar × Shlok Khemani × Tom McGrath × Andrew Moore

Fable’s usable autonomy depends on interface gates: it independently combined satellite imagery with NASA elevation data, yet production access often triggered a fallback to Opus 4.8.Hybrid authorship is gaining traction as Frontier Code’s merge acceptance rose from roughly 10% to 25% and upwards of 30%, but Mythos’s research evidence still trails its engineering acceleration while reward hacking and illegible reasoning keep alignment unresolved.

Erik Torenberg × Nathan Labenz × Dan Balsam × Tom McGrath

Goodfire’s $150 million Series B at a $1.25 billion valuation turns its interpretability research into an execution test spanning seven-figure engagements and a scalable AI stack.Its hallucination experiment used a frozen reward model to improve behavior without obvious capability loss, but robustness remains conditional on probe quality, training configuration, and scale.

Daniel Balsam × Tom McGrath

Mechanistic interpretability now has the raw materials of a “proto-paradigm”: understandable features, linear directions, superposition, and circuits, despite lacking consensus.Existing tools already help customers, with scientific discovery and inference-time guardrails emerging as important applications while algorithms remain the softer bottleneck.Yet a scaling study found reconstruction curves bending before 99.99% or 100%, leaving memorization, higher-order geometry, or noise unresolved as Goodfire advances its $50 million independent category bet.