Sam Altman clarifies AI cost remarks, says low-cost LLMs are now ‘doable’
Back in 2023, Altman had said that it was "totally hopeless" for smaller teams with limited budgets to compete with OpenAI on training foundation models, a remark that sparked considerable debate in the AI community.
ADVERTISEMENT
OpenAI CEO Sam Altman, on his second visit to India on February 5, clarified that his earlier comments about the high costs of developing frontier AI models in India were taken out of context.
Speaking at a fireside chat with IT Minister Ashwini Vaishnaw , Altman said "That's a reference to a comment I made here a few years ago about cost. Maybe that was taken out of context. That was at a very specific time when there was a certain scaling thing, where I thought, and I still think, to stay on frontier of pre-trained models, is expensive," Altman said.
Back in 2023, Altman had said that it was "totally hopeless" for smaller teams with limited budgets to compete with OpenAI on training foundation models, a remark that sparked considerable debate in the AI community.
Sam Altman's comments come at a time when there is debating raging, especially after the success of DeepSeek, over whether creating a LLM should require billions of dollar in capital and access to advanced semiconductor chips. DeepSeek had reported comparable results at a fraction of cost with less-advanced chips from Nvidia than what the Western world had been working with.
However, in his latest comments, Altman shifted focus to the progress that has been made in AI model development. "One of the most exciting things that's happened to me since, that I can tell in the industry since, is we're now in a world where we made incredible progress with distillation," Altman said."We learned a lot to do with small models. And these reasoning models in particular can be, it's not cheap, it's still expensive to train them, but it's doable," he added. Altman said that "we can look at costs in two different ways-- these costs will continue to rise on exponential curve, but the returns from increasing intelligence will also be exponential in terms of the economic value and the scientific value that you can create."
"What's happening with the reduction in the cost of AI models is extraordinary. Now, I don't think it means the world will need less AI hardware. As you bring the cost down, people are simply going to use it for many more things," he added.
IT Minister Vaishnaw added that India's entrepreneurial talent is really focused on getting the next level of innovation 'that will reduce the cost', and hoped to raise the bar on cost-efficient LLMs. "On the same lines that we did the Chandrayaan mission, why can't we do the same in the LLM space," asked Vaishnaw.
