Fifteen Months Before ChatGPT

In September 2021 a Columbia University computer science professor put a commercial product built on GPT-3 into live production. Not a demo, not a paper: a working application answering queries on ESPNcricinfo, the sports site he had co-founded as a graduate student and which ESPN had bought in 2007.

ChatGPT would not exist for another fifteen months. There was no consensus that these systems were useful, no market to be early to, and no accepted theory of why they worked at all.

Vishal Misra, now RKS Family Professor of Computer Science and Vice Dean of Computing and AI at Columbia Engineering, built the thing first and went looking for the explanation afterwards.

The Theory Came Second

Academics do not usually work in that order, and he says so plainly. "I backed into doing research on LLMs (Large Language Models) like ChatGPT." The conventional sequence, in his description, is fundamental research first, a commercial application spotted downstream, then an attempt to commercialise it.

He ran it backwards. "In my case, I first built a commercial application using GPT3 which has been running in production on ESPNCricinfo (a startup I co-founded when I was a grad student) since September 2021 - about 15 months before ChatGPT was released." Only then, "playing with GPT3", did the question of how it worked take hold.

The answer became a research programme. With Professor Sid Dalal of Columbia's statistics department, Misra says the pair have "a model that mathematically explains how and why these LLMs work, what their strengths and limitations are, and how they can help with research."

That is an unusually large claim, and it is worth noting what it rests on: a product that had already been running for two years by the time he made it.

What He Tells His Students

In 2024, Dean Shih-Fu Chang created a position that had not previously existed at Columbia Engineering - Vice Dean of Computing and AI - and made Misra its first holder, with a brief covering computing and AI across teaching and research, and the ethics of both.

The pitch he makes for the technology is about access rather than capability. Deep learning had been improving since around 2011, he notes, "but ChatGPT was the first time everyone could directly use AI - from a middle school student to a lawyer to a retiree." The consequence: "You don't have to be tech-savvy to use cutting-edge AI anymore."

Then the qualification, in the same breath. "But just because you can use it doesn't mean you know how or when to use it."

His working analogy for what he is handing students is not flattering to the tools. He describes ChatGPT as "the most helpful and knowledgeable entity you can hope to encounter, but with early onset dementia!" - a formulation that puts the burden squarely on the user. "You really need to verify what it confidently asserts", he says; hallucinations, he adds, "are a big problem."

An AI vice dean whose headline advice is do not trust the output is not the usual posture for the job.

The University Cannot Buy Compute

The complication in his brief is not intellectual. It is a procurement problem, and he has stated it in public.

Columbia's research community does not have the machines. Misra's stated aim is to "ensure that Columbia's research community has similar compute capabilities available like the big tech companies (Google, Microsoft, Amazon, etc.) or companies like Open AI" - a sentence that concedes the gap in the act of describing the goal.

Two routes are on the record. The first is Empire AI, a consortium between the State of New York and six universities, whose groundwork he credits to Executive Vice President for Research Jeannette Wing and Dean Chang. The second is barter: the school is "in active discussions with major cloud providers", offering research output and utilisation in exchange for access to infrastructure it cannot otherwise afford.

The candour extends to the institutional politics. "AI has hit all of us right in the face," he says, arguing that the computer science department and the engineering school are the natural leaders in getting a whole university to use the tools properly. Natural leaders is a claim, not a fact, and every dean at Columbia knows it.

A Board Seat In Construction

On April 22, 2026, Procore Technologies - the NYSE-listed construction management platform on which more than three million projects have run across over 150 countries - appointed Misra to its board of directors.

It is a specific kind of appointment. Procore's president and chief executive Ajei Gopal framed it around what the company is trying to build, saying Misra's "deep expertise in AI and proven success as a technology founder will be a tremendous asset."

Misra's own framing put the money where the buildings are. Procore, he said, "is at the forefront of AI innovation in construction, driving a fundamental shift in one of the world's most essential industries." The reason to be there now: "The surging demand for critical infrastructure presents an unprecedented market opportunity."

The pattern is consistent across thirty years. He co-founded Cricinfo and is credited with inventing live-microblogging there. He founded Infinio in storage performance and AskHereFirst in natural language query. He took a B.Tech from the Indian Institute of Technology, Bombay in 1992 and a doctorate from the University of Massachusetts Amherst in 2000, is a Fellow of both the ACM and the IEEE, and past chair of ACM SIGMETRICS. In every case the applied work and the formal work have run alongside each other rather than in sequence.

Which leaves the open item. He has a mathematical model of why large language models work, a school-wide mandate to teach a university how to use them, a board seat at a company deploying them into construction sites, and not enough compute to run the research that would settle any of it. The consortium is assembled. The cloud negotiations are, by his own account, still discussions.