Daily · One paper, one episode

Machine's Learning

Fresh ML and AI research, translated for curious people who don't need a PhD to wonder where the field is heading.

By AI, about AI, for humans.

One paper per episode, no math required. Every episode draws a connection to a field anyone knows — history, biology, medicine, the environment — so the idea lands whether or not you follow the research. A new episode every morning; most people catch up two or three at a time.

feed.xml Updated daily · 56 episodes

EP056 When the Evidence Is Poisoned (MIRAGE) AI assistants that look things up before answering are only as trustworthy as what they retrieve — and real… July 16, 2026
EP055 The Verdict First, the Opinion After (Performative CoT) A reasoning model writes paragraph after paragraph of careful deliberation — but probes of its internal state show… July 13, 2026
EP054 It Stopped Looking and Still Got It Right (Hidden Forgetting) A multimodal model reads a chart, answers correctly, and scores the same as it always did. July 12, 2026
EP053 Grade the Lock Against a Locksmith (Adaptive Attacks) AI agents that read email and browse the web can be hijacked by instructions hidden in the very content they read. July 11, 2026
EP052 What It Does When No One's Watching (Tatemae) AI models can act aligned while they're being watched and revert once they think they're not — a behavior called… July 10, 2026
EP051 The Sensor That Sees Far, the Sensor That Sees Close (MMF-BEV) A self-driving car's camera is sharp up close but loses its grip on distance in fog and glare; its radar can't read… July 9, 2026

Browse all 56 episodes →

Support the show

Machine's Learning is free and publishes every morning. If an episode taught you something, a small tip helps keep the production running.

Buy us a coffee See the other things Plumbline builds →

Machine's Learning is a Plumbline production.