The afternoon that changed Craig S. Smith’s beat

Craig S. Smith went looking for a story about Toronto’s technology scene and came away with a different career. The assignment led him to spend an afternoon with Geoffrey Hinton, who introduced him to neural networks. Smith wrote his profile for The New York Times. Then he kept going. An interview that might ordinarily have ended with publication had opened a subject he wanted to spend considerably more time inside.

For a foreign correspondent, the sequence was familiar enough: arrive somewhere, find a knowledgeable person, discover how much there is to learn. This time, the unfamiliar territory was a field of research. There were conferences to attend, courses to take and researchers to question. He had spent years moving between countries. Now he was learning his way around a different kind of map.

The result was Eye on AI, the publication and podcast he founded in 2018. Its origin was unusually practical. Smith wanted to keep interviewing senior researchers, and those conversations did not have a ready home at the Times. So he made one. The audience came later, he has recalled. Curiosity had already booked the appointment.

A publisher meets the firewall

Long before he hosted conversations about machine intelligence, Smith had worked through the demands of foreign reporting. He wrote for The Wall Street Journal before joining the Times as Shanghai bureau chief in 2000. Paris followed in 2002. His reporting took him to more than 40 countries, including assignments covering the wars in Afghanistan and Iraq and the 2006 conflict between Israel and Lebanon.

That experience gives the AI chapter a particular backdrop. A correspondent must work out who understands a situation, what interests they represent and what their account leaves unresolved. Technical reporting presents its own version of those questions. A demonstration can be impressive while the conditions required to reproduce it remain a story in themselves. The habit of asking another question travels rather well.

Smith also spent time on the business side of information. In 2008, he joined Richard Li’s financial news venture as executive editor, later becoming a senior vice president at Pacific Century Group. He returned to the Times in late 2011 as China managing director. His responsibilities included founding and running its Chinese-language website.

The site launched in June 2012. By October, it was blocked in mainland China after publishing reporting on the wealth of the family of Wen Jiabao, China’s former prime minister. In a 2013 interview, Smith explained the mechanics of internet restrictions and the ways private platforms were made responsible for policing content. Publishing involved an additional question: could the intended reader actually reach the work?

He described strong early growth and demand from luxury advertisers before access was restricted. The operation also launched a Chinese-language lifestyle site. This was publishing with the distribution problem very much attached. An audience could want the journalism, an advertiser could want the audience, and a blocked domain could interrupt the entire arrangement. A spreadsheet has limited diplomatic powers.

A new beat needs a place to live

Smith returned to North America in 2016 to work on Canadian stories. The Hinton encounter followed in 2017. By 2018, he had retired from the Times, although he continued writing for it and other publications. Retirement, in this case, left considerable room for deadlines.

His move into independent media brought together two parts of his working life. He knew how to conduct an interview, and he had experience building the operation that would carry it. Eye on AI gave him editorial room to stay with a subject beyond the point at which an individual newspaper assignment would normally end.

The publication’s stated offering includes original articles, newsletters, interviews, aggregated news and event coverage, aimed at industry professionals and the financing community. Its team page names Smith as CEO and financial journalist Tim LeeMaster as executive editor. It also credits the people behind the site’s design and music. Even a conversation with one voice asking the questions has a production life around it.

The podcast describes its purpose as placing individual advances in a broader context and considering their global implications. That is a useful editorial promise because a new result and an explanation of its consequences are different pieces of work. The first can arrive as an announcement. The second requires someone to stay on the telephone.

“I did not consider an audience at that time, only my own interest.”Craig S. Smith, reflecting on the podcast’s beginnings

What the guest list reveals

The early archive offers a compact guide to Smith’s interests. Richard Sutton appears in a conversation about reinforcement learning. Samy and Yoshua Bengio discuss their intertwined professional lives. Yann LeCun talks about self-supervised learning and the route from his early interest in computers to his research. These are people with histories as well as ideas.

Other episodes move outward. Benjamin Rosman discusses machine learning in Africa. Partha Talukdar addresses the field in India. Eric Schmidt and Bob Work consider national security and engagement with China. Climate researchers David Rolnick and Priya Donti appear in a conversation about organizing machine-learning work around a warming world. Geography remains part of the beat.

A guest list is a series of choices about what deserves attention. Here, the technical questions sit beside questions of institutions, national capacity and the uses to which research might be put. That range fits a reporter whose earlier work required him to look across borders, even when the immediate subject was close at hand.

Smith’s involvement in AI policy extended beyond interviews. He served as a special government employee for the National Security Commission on Artificial Intelligence until the commission ended in October 2021. That chapter belongs alongside his reporting and publishing work when considering how he came to know the field.

There is a practical side to the technical coverage, too. In a series with Determined AI co-founder Ameet Talwalkar, Smith introduced conversations about the machine-learning pipeline: preparing data, developing models, training them, managing hardware and deployment. The series began with David Patterson. It is the sort of itinerary that follows the work through its machinery, giving the listener more to inspect than a finished demonstration.

The pencil beside the microphone

The professional biography is busy, but a few smaller details make Smith easier to picture. Asked about a nontechnical hobby in January 2023, he said he made hand-drawn animations. Asked about a guilty pleasure, he named TikTok. A person can spend his day investigating machine intelligence and still surrender a portion of it to the scroll.

He also described writing every day and using podcast conversations as inspiration for his articles. The difficulty he singled out was finding qualified people to comment on the technology. It is a revealing complaint in a field where commentary itself is plentiful. Access to someone who can explain a particular piece of work remains valuable, however crowded the conversation becomes.

In that same interview, Smith said he was considering a master’s program in computer science. The statement captures an appetite for further study, without needing a ceremonial story about reinvention. Someone who has spent years asking specialists to explain their work can still want a more formal way into the subject. Experience does not exhaust the supply of things to learn.

His published work at IEEE Spectrum gives that interest a concrete outlet. His articles have examined language-model hallucinations, AI-assisted video compression and China’s Wu Dao models. Those subjects require attention to limits, infrastructure and the decisions behind a product. They also offer readers a way to follow the technology beyond the names of the companies selling it.

The next questions are about action

The beat continues to change. In a December 2024 interview, Smith identified agentic workflows and groups of cooperating agents as subjects he wanted to explore. He also pointed toward world models that learn through interaction with the physical world. These were his stated interests and expectations, rather than a timetable readers could put in a calendar.

His more recent publication work includes Eye on AI Weekly Research Watch. The Substack archive lists an August 2026 edition offering twenty short audio explorations of arXiv papers and an October edition of the research digest. The format extends his original interest in conversations into another recurring task: choosing research that deserves a closer look.

Research coverage has a different rhythm from a product launch. A paper can raise a question before there is a familiar object to photograph or an application to try. Following that work asks readers to spend time with possibilities still being tested. It gives a publication a reason to return to a subject after the first headline has moved on.

Smith’s career now contains both kinds of movement: the long geographical journey of a foreign correspondent and the repeated, smaller journey into an unfamiliar idea. The afternoon with Hinton connected them. A reporter arrived with an assignment, heard something that held his attention, and arranged to keep learning. Several hundred conversations later, that remains a perfectly serviceable reason to turn on a microphone.