Fall 2025
By Alice Levitt
“We live in an era where you can get a Nobel Prize in four years,” said Anshumali Shrivastava.
A professor of computer science, electrical and computer engineering, and statistics at Rice, Shrivastava is also a co-founder of ThirdAI with former Rice Ph.D. student Tharun Medini. “He’s the smartest kid you’ll find,” Shrivastava said.
A future Nobel laureate? Chances look good if the four-year success of the AlphaFold artificial intelligence model is anything to go by. ThirdAI (pronounced “Third Eye,” referencing perception beyond ordinary sight) is moving with similar speed. Formed in 2021, Shrivastava said, “There was no generative AI, there was no ChatGPT” at the time of its inception.
“We knew that an age is going to come where people will need very large neural networks, very large models with a lot of data,” he recalled. “We knew it was coming in some form or the other … so we started building an infrastructure where we were looking at the science and technology of how we can create AI that is 100 or 1,000 times more efficient than what it was [at the time].”
What the pair built was uniquely efficient AI that makes large-scale modeling practical. Most AI relies on graphics processing units for power. ThirdAI focuses on using central processing units — what most home computers run on. For countries that can’t afford GPUs, this is critical.
Is this already going over your head? Shrivastava forgives you. “Everybody only has a shallow expertise because it is so new. So we are learning,” he said. “A lot of people call themselves an expert in LLMs, but how can you be an expert when a field started two years back?”
LLMs — large language models — have been the talk of the tech world. This spring, ThirdAI got major attention on Hacker News after releasing a method for lossless compression of LLMs. “When running LLMs efficiently, everybody has to make a compromise, which typically comes in terms of accuracy,” Shrivastava explained. “And right now, people are really worried because they spend a fortune to get their accuracies high, and people are like, ‘Maybe the technology is not there.’”
But his team’s breakthrough proves that not only is the tech there, but it can also reduce costs by as much as 50%.
All you need is yourself and a computer, and you can literally change the world.
Even before that leap, ThirdAI had made major strides at Rice. Tam Dao, associate vice president of campus safety and research security, works to protect Rice research from foreign espionage. Working with Dao, ThirdAI created Preventive RISk Monitoring (PRISM), a program that cut down assessments from 6 to 8 hours to 2 minutes.
“His team and our team meet every Monday, and they essentially just replicate what I would do manually into an automated process that’s quicker, much more reliable and much more valid than I would be able to do by myself,” Dao said.
This shift frees up his team to focus on prevention rather than intervention, making for a safer campus. And by all accounts, the ThirdAI team is also a pleasure to work with.
“I’ll tell you, honestly, I don’t understand 1% of what they really do,” Dao admitted. “You know how sometimes you can work with really smart people, but they just can’t connect to you? … These guys are on different levels in terms of intelligence, but they also have the social skills to work with me. … It’s been an incredible pleasure.”
At a time when the world is still grappling with the place of AI in our lives, ThirdAI is already putting it to work. Shrivastava said that many business owners still struggle to understand what AI is actually good for.
“But you know what the bigger problem is?” he asked. “We don’t even know what to do with AI. … Then you figure out, OK, if 70%–80% of the world is there, should I wait for the world to come to a conclusion … or should we go and solve who has a problem?”
Learning how to match products to real needs has made Shrivastava’s research more effective. “It’s very easy, as a researcher, to live in a bubble that you think is the world’s problem — to just take a hypothesis and validate it with some numbers, and then you think that’s the real world,” he said.
But the real world is paying attention. In regions where GPUs are cost-prohibitive — like parts of Eastern Europe, India and a large African nation — ThirdAI’s model is generating interest.
India is especially personal for Shrivastava and Medini, who both hail from there. The company’s name borrows from Eastern philosophy, and its 16-person team maintains a small office in Bengaluru, where the Rice Global India campus opened in 2024 (see Page 105, “Global Ambitions Take Root in India”).
“There is a very nice ecosystem of engineers there, and they are very hard-working. … We understand that ecosystem. We know where to get some of the smartest folks,” said Shrivastava, a graduate of the Indian Institute of Technology.
He also credits Houston with helping to shape ThirdAI. Most of his business partnerships are abroad or in Silicon Valley, but when it comes to hiring, he said the “vibe” of Houston attracts talent, including several Rice grads beyond Medini.
By the time you read this, it may already be time to update ThirdAI’s story. And that’s what keeps Shrivastava showing up each day.
“All you need is yourself and a computer, and you can literally change the world,” he said.