Publishing Details
Contact & Outreach
About This Podcast
Explore Statistics
Recent Episodes
EP038 — Presence Is Provable, Absence Is Not (Orca PU Learning)
Underwater microphones have listened to one stretch of the Pacific for more than thirty years — and almost none of it is whale. From that mostly-empty ocean, researchers assembled the largest…
EP037 — Run It Until It Stops Moving (Attractor Models)
Some AI models think by running the same small step over and over, polishing an answer each pass — but you have to pick the number of passes in advance, and training them is unstable and expensive. A…
EP033 — One Yardstick Makes a Monoculture (Model Collapse)
As more of the internet becomes AI-generated, future models will be trained on the output of past ones — and the fear, called model collapse, is that quality decays generation by generation like a…
EP032 — Say the Hard Part Out Loud (Selective Latent Thinking)
Reasoning models get more accurate by "thinking out loud" — writing out their working step by step — but every word costs time and compute. This paper teaches a model to say only the hard parts out…
EP031 — How Much Of It Is Real Belief (Collective Alignment)
Solomon Asch's 1951 line-judgment experiment gave social psychology its first apparatus for separating two things that look identical from the outside: what a person would perceive alone, and what…
EP030 — The Red Car That Becomes a Truck (ITC World Models)
An AI agent that learns a little internal model of how its environment behaves can practice inside that model — predicting what the next video frame will look like, then the next, then the next. A…
EP029 — Teaching an Agent to Find the Wires (MechRL)
Inside a small open-source language model called GPT-2 there are 144 attention heads, and for any specific task — say, completing a sentence so the right name fills the blank — only a small handful…
EP028 — A Benchmark That Doesn't Travel (Camera-Trap Drift)
A new unified benchmark across 546 camera traps reveals something striking: BioCLIP 2, the current biological foundation model for ecological vision tasks, underperforms at numerous sites even in the…
EP027 — Each Parameter, Its Own Place (KAN-CL)
Catastrophic forgetting is the central obstacle when a neural network has to keep learning over time — train it on a new task and its competence on the old one collapses. The field's standard fix has…
EP026 — When Forgetting Is the Point (Human-Inspired Memory)
Most AI agents store every observation in a vector database forever and call that memory. Doga Kerestecioglu and colleagues at Microsoft (arXiv 2605.08538) build the opposite — a memory system openly…
EP025 — Safe Parts, Unsafe Machine (Interaction Topology)
Take four AI agents, each one individually safe and well-behaved, wire them into a committee that decides together — and the committee can do things none of them would. A new position paper from…
EP024 — One Task, Four Hidden Programs (ICL Phases)
In-context learning — a transformer picking up a pattern from examples in its prompt without any change to its weights — is usually treated as a single capability a model either has or doesn't. A new…
EP023 — The Model That Never Heard a Whale (Perch 2.0)
Perch 2.0 is a bioacoustic AI trained on the sounds of 14,597 mostly land animals — birds above all — recorded in open air. It heard almost no marine mammals. Yet when a team at Google DeepMind froze…
EP022 — When One Imagined Future Isn't Enough (Probabilistic Dreaming)
An AI agent that learns a little internal model of how its environment behaves can practice inside that model — the field calls it dreaming — instead of acting things out in the real world. A new…
EP021 — Measuring What Capacity Has Left (Predict Plasticity)
After enough training on a long sequence of tasks, neural networks quietly lose the ability to learn anything new — the field calls it loss of plasticity, and it has a string of standard…
EP020 — When the Number Isn't the Truth (Camera-Trap Fusion)
Camera traps scattered across North Carolina are generating images by the millions, and the bottleneck is no longer the cameras — it's labeling. An AI classifier can label every image, but the number…
EP019 — Three Ways to Watch for Deception (DeceptGuard)
Imagine watching a contractor remodel your kitchen and being able to peel back three layers of access: first only what they do, then their reasoning out loud, then a recorder that captures every…
EP018 — Mapping the Patterns at Scale (Attention Atlas)
Most interpretability research zooms into one attention head, in one model, on one prompt — careful, slow, by hand. A new paper from Jonathan Katzy, Razvan-Mihai Popescu, Erik Mekkes, Arie van…
EP017 — A Query With No Words (LAnR)
Almost every AI chatbot you have used in the last two years has a look-up step underneath — retrieval-augmented generation, or RAG. A new paper from Ha Lan N.T, Minh-Anh Nguyen, and Dung D. Le at…
EP016 — When New Learning Erases the Old (FTN)
Train a neural network on task A, then on task B, and its competence on A typically collapses — the field's longest-running open problem in continual learning, called catastrophic forgetting. A new…
Frequently Asked Questions
Machine's Learning has published 35 episodes since April 2026, covering topics in Technology.
Machine's Learning is currently active with new episodes daily. Average episode length is 17m.
Sign up on Grep.FM to access contact details for Machine's Learning, including email and social media links.
