In a brief onstage conversation with IT Minister Ashwini Vaishnaw, OpenAI CEO Sam Altman said that his remarks in 2022 that it was “hopeless” for Indian firms to compete with the United States on building a foundational Artificial Intelligence model were “taken out of context”. Those remarks have been in focus since the Chinese firm DeepSeek managed to build a foundational model at par with OpenAI and other leading firms’ AI models at a fraction of the cost. OpenAI was “really excited to do a lot more together” in India, Mr. Altman said.
Mr. Altman cited so-called small language models, which are less expensive to build, and also acknowledged the decreases in cost that training AI models had undergone. “It’s still expensive to train” advanced AI models, Mr. Altman conceded, adding that “to stay at the frontier, costs will rise on an exponential curve, but also the returns to increase in intelligence are exponential in terms of the economic value, the scientific value that you can create.”
The cost of one “unit of intelligence,” Mr. Altman said, is falling tenfold annually. “Our young entrepreneurs and our startups are really, really focused on getting that next level of [cost reduction],” Mr. Vaishnaw said, citing the Chandrayaan Mission’s low cost. “Of course, we are creating certain very good data sets which will be used for training the models in an Indian context in terms of language and the cultural and regional nuances within the country,” Mr. Vaishnaw said.
Mr. Altman also spoke of the firm’s new deep research model, and hoped to make a lot of intellectually taxing work far more efficient, especially in research fields. “If you are a scientist trying to cure some disease, deep research is surely not going to cure that disease on its own,” Mr. Altman said. “But if you can farm out the tasks that took you a lot of time but were lower value” like literature reviews or equipment ordering instructions, “maybe you can be twice as efficient.”
Published – February 06, 2025 02:13 am IST