Eye on AI asks
the next question

In October 2018, a new podcast feed carried two demo conversations with Jack Clark, then working on strategy and communications at OpenAI. The subjects included international rules for artificial intelligence, access to computing power, and whether machines might become a mirror of human cognition. Nobody needed a prompt cheat sheet. The question was bigger: who would get to shape this thing?

The interviewer was Craig S. Smith, a longtime New York Times correspondent. His company, Eye on AI, has kept that microphone open through successive waves of enthusiasm. Scroll backward through its archive and today's arguments acquire ancestors. Computing inequality, machine reasoning and military applications were already there. The vocabulary changes faster than the underlying questions.

The reporter in the research lab

Smith's background helps explain the product. Eye on AI's team page identifies him as CEO and credits him with building the Times' Chinese-language platforms. It also names executive editor Tim LeeMaster, whose experience includes financial journalism in Asia and equity capital markets coverage at Acuris. These are people accustomed to asking what a development means outside the room where it happened.

Their stated audience is industry professionals and the financing community. The problem they address is familiar to anyone who has opened a research paper after reading a triumphant headline: the distance between a technical result and a useful understanding of it. Eye on AI supplies interviews, original articles and research digests that give readers more ways into that gap.

Consider the guest list. Richard Sutton appears in episode 11 to discuss reinforcement learning. Geoffrey Hinton appears in episode 63 to discuss learning in the brain. Episode 111 revisits conversations with Terry Sejnowski, Hinton, Yann LeCun and Andrew Ng. The attraction is hearing researchers explain the ideas behind systems that other people encounter as finished products.

Yesterday’s ambition, today’s checklist

One especially useful archive entry is episode 33. Justin Gottschlich, who founded Intel Labs' machine programming research group, describes efforts to let people create software by explaining what they intend it to do. Read that description now and it sounds remarkably contemporary. Its value is historical perspective: an apparent overnight arrival can have a long technical prehistory.

Other episodes reveal competing routes. David Cox's discussion of neuro-symbolic AI explores combining logic-based approaches with deep learning. Pedro Domingos returns in episode 250 with an argument for unifying five machine-learning traditions. Eye on AI lets those disagreements coexist in the catalog. A listener can compare explanations instead of treating the most recent announcement as the field's settled position.

That suggests a practical way to use the product. Pick a question your team actually faces, find an interview that addresses it, then write down the mechanism, the evidence and the assumptions. Follow the transcript when a term gets slippery. Bring the resulting questions to a vendor or researcher. The payoff is a more informed conversation; the episode itself cannot validate your deployment.

Two inboxes, one research habit

The subscription form offers AI Finance Watch and AI Research Watch. The Research Watch archive is particularly concrete. Its week-ending October 4, 2026 edition includes summaries of newly published papers, author names, dates and links. One covers geometric representation learning from novel view synthesis; another examines agents that reconstruct moving scenes as executable graphics programs. Those are useful doors into work that rarely fits a tidy product announcement.

The rhythm matters. An interview provides a person's explanation; a digest supplies things to investigate afterward. Audio suits a commute, a transcript suits a search, and the paper link suits a closer examination. This is an editorial product with several reading speeds. For a researcher, investor or product manager, the sensible habit is to move between them rather than confuse a summary with the underlying result.

“AI is about to change your world, so pay attention.”Eye on AI · podcast description

The audience is also the inventory

Eye on AI's commercial offer is sponsorship. Its contact page solicits newsletter and podcast sponsors and describes prominent branding in newsletters, audio and program notes in English and Chinese. The archive makes the arrangement tangible: episode 23 labels its Determined AI conversation as the first in a periodic sponsored series; episode 26 labels its Labelbox conversation as the second.

A 2024 Pedro Domingos episode names Bloomreach as its sponsor. These are documented commercial relationships, not evidence that every guest is a customer. For advertisers, the offer is proximity to an audience interested in AI research and business. For listeners, sponsorship is useful context when assessing a company's explanation of its own technology.

The public episodes are available through the website and podcast platforms without a displayed episode purchase price. The cost a listener can measure is time: Sutton's interview runs 38 minutes and eight seconds; the Domingos episode 250 runs an hour, eight minutes and twelve seconds. That is a substantial appointment with an idea. Listening selectively beats completing the catalog as homework.

Keep the second question

Eye on AI occupies the specialist AI media market. An alternative such as Sam Charrington's TWIML AI Podcast also interviews people working in machine learning. Eye on AI's particular appeal is Smith's reporting background, its connection between research and finance, and an archive stretching back to 2018. That combination rewards readers who want continuity as well as updates.

The feed remains active. On October 6, 2026, Smith interviewed Front CEO Dan O'Connell about customer operations and the coordination work surrounding AI. The episode description draws on a survey of 700 leaders and challenges the assumption that automation straightforwardly reduces organizational effort. It is a fitting subject for a publication built around asking what happens after the promise.

The method is easy to copy: preserve earlier predictions, return to unresolved questions, and ask practitioners to explain the work between demonstration and result. Interviews are less useful when you need audited performance, independent procurement advice or instructions tailored to a specific system. Use them to sharpen an investigation. Then ask the next question yourself.