How this Houston innovator is making AI accessible, personal, and safe

HOUSTON INNOVATORS PODCAST EPISODE 198

Anshumali Shrivastava joins the Houston Innovators Podcast to share the revolutionary work ThirdAI is doing for artificial intelligence. Photo via rice.edu

Anshumali Shrivastava's career has evolved alongside the rise of artificial intelligence. Now, he believes his company represents the future of the industry's widespread implementation.

Shrivastava, who's also a professor at Rice University, founded ThirdAI, pronounced "third eye," in 2021 to democratize artificial intelligence through software innovations. As Shrivastava explains on the Houston Innovators Podcast, AI processes have historically been run on larger, less accessible computing hardware. ThirdAI's tools are able to run on a regular central processing unit, or CPU, rather than the more powerful graphics processing unit, or GPU.

"We focus on the problems that people are facing in the current AI ecosystem," Shrivastava says on the podcast. "Right now, if you are to build some of the large-language models and (linear programming) models, you need a lot of computing power, dedicated engineers to move it, and, even if you are using fully managed services, it's costly and there are a lot of privacy implications because you have to move data around."



These are some of the challenges for AI development, and, as Shrivastava points out, this process isn't even accessible for 90 percent of the world that lacks the infrastructure to do it. And, even for companies that can afford to invest in the dedicated GPU hardware, there's a global chip shortage.

ThirdAI's solution? Enable AI processes on the hardware that is accessible — CPUs.

"That is what our product offerings are — the AI ecosystem on commodity infrastructure," he says, explaining that ThirdAI's goal is also to improve existing AI applications.

One specific AI application that ThirdAI is making more effective for its customers is search tools in ecommerce. The need to make online shopping searches as quick and as accurate as possible directly affects the company's ability to complete the transaction. Wayfair tapped into ThirdAI's tech to address its latency in its on-site searching.

"Their problem was in the domain of making language models and search engines, which are AI based, very efficient," Shrivastava says. "We were able to bring down (latency) significantly with our technology."

One AI application that's taken off over the past few years is the chat-based model — led by OpenAI's ChatGPT. As exciting as the prospect of navigating information via chatbot is, many companies, understandably, have privacy concerns.

ThirdAI created PocketLLM to address this concern. The platform, which ThirdAI offers for free, operates completely on the harddrive of the owner of the data, meaning your data stays with you.

"ChatGPT has shown the world what is possible," Shrivastava says, explaining that 80 or 90 percent of use cases are people or companies wanting to take their knowledge and data and turn it into an AI chat tool. "What people want is a ChatGPT-kind of agent on their data, but they don't want their proprietary data to be leaked to the outer world."

"Because I can build AI where your data is, your data never have to leave your ecosystem. PocketLLM is a demonstration of that capability," Shrivastava continues. "It's a tool that uses our software stack and essentially looks at your data and builds this chat agent and offers you an air-gapped privacy."

Shrivastava explains that the tool can be used for enterprises or even personal applications — like navigating past conversations on email, seeing as no email provider seems to have an optimized search option.

PocketLLM is just one thing ThirdAI is working on. Shrivastava explains that the company is developing an entire ecosystem of tools that can be used on CPUs.

"If AI is going to be an agent, it better be personalized," he says, explaining that no one AI will have the right answer for everyone.

Shrivastava shares more about the future of both AI and ThirdAI, which is growing to keep up with demand. Listen to the interview here — or wherever you stream your podcasts — and subscribe for weekly episodes.

From Your Site Articles
ThirdAI's new PocketLLM app is free to use and completely secure. Photo via Getty Images

Houston startup launches innovative chat tool on its mission to democratize AI

smart tech

Artificial intelligence has a big potential to disrupt the technology industry, and one Houston company that was founded by a computer science professor at Rice University, is fast on its way to help lead that future now in a convenient and affordable way.

Founded by Anshumali Shrivastava and Tharun Medini, a recent Ph.D. who graduated under Shrivastava from Rice's Department of Electrical and Computer Engineering, ThirdAI is building AI deep learning tools that aim to be sustainable and scalable to fit the changing needs of the industry. The company is on a mission to democratize AI, Shrivastava tells InnovationMap.

Shrivastava likes to use the word efficiently when describing what makes ThirdAI different, and how its programs can teach AI via multiple avenues to be what he refers to as “1,000 times more efficient.”

“The carbon footprint of these models are off the charts, and so expensive,” Shrivastava. “We believe this could be made efficient. … We use the same ideas that were developed, but we do it on a massive scale.”

ThirdAI's latest tool is a multilingual ChatGPT-like AI training tool PocketLLM app. Announced earlier this month, the tool is free. According to the company, users have access to a personalized chatbot that understands what the user is searching within documents, and can be fine-tuned to help elaborate your thoughts through a neural search.

ThirdAI's PocketLLM app is free to use. Image courtesy of ThirdAI

The app is private and secure and runs on deep-learning algorithms according to Vinod Iyengar, head of product at ThirdAI, and no one — not even ThirdAI — has access to the documents except the user.

“Tools exist to help people search text files, but that requires sharing your data with third parties,” says Iyengar in a news release. “Our solution is private and secure, powered by deep learning algorithms. And it returns results lightning fast.”

The process includes the user installing the app, uploading any text document files, and clicking "train." Minutes later, you have an AI tool that can process the information in those documents.

“The neural search encourages you to elaborate on your thoughts with details in the discover window and see the difference in results,” says Shrivastava in the release. “It can also be fine-tuned to your tastes by selecting the relevant option and hitting the update button to re-train."

In September of 2021, ThirdAI — pronounced "third eye" — raised $6 million in seed funding. The round was invested in by three California-based VCs — Neotribe Ventures and Cervin Ventures, which co-led the round with support from Firebolt Ventures. The technology ThirdAI is working with comes from 10 years of deep learning research and innovation. The company's technology has the potential to make computing 15-times faster, the company reports.

Anshumali Shrivastava is an associate professor of computer science at Rice University. Photo via rice.edu

Ad Placement 300x100
Ad Placement 300x600

CultureMap Emails are Awesome

Property Showcase

Houston lab explores how AI bots can help the elderly

AI for aging

The University of Houston’s Empathetic Lifespan AI & Robotics for Aging (ELARA) Lab is currently conducting research into how AI bots may be able to help the elderly live more social and independent lives through several ongoing initiatives.

The lab officially launched last month as part of the Gerald D. Hines College of Architecture & Design under the leadership of Assistant Professor Chorong Park. Part of the lab’s mission is tackling ongoing problems with aging, such as dealing with disabilities and social isolation. Researchers’ current work is focused on designing a new AI companion bot specifically tailored to the needs of older people.

“We need to take all the needs of older adults seriously,” Park said in a news release. “They won't use the robot if they don't feel at ease or if they feel they are being constantly watched.”

The field testing of new AI bots in this population hopes to overcome several traditional obstacles in technology use among the elderly. A study by Park shows that many older people have a fear of overt surveillance when using advanced AI. There is also ageism to consider. Most new technologies are designed with younger and employed buyers in mind, not retirees who may need help remembering daily tasks or accessing important information.

“The more older adults are excluded from technology development, the worse those technology gaps will become,” Park said. “AI and the majority of technologies are created for younger people, so my research method integrates older adults directly into the design process.”

ELARA recently collaborated with the Mamie George Community Center in Richmond, Texas, to track seniors’ response to desktop AI bots like Emo and Cupboo. Researchers also had participants use air-dry modeling clay to create their ideal robotic companion.

While the eventual AI bot may be able to help the elderly feel less isolated and more supported, there are concerns to consider. A study published in the Asian Journal of Psychology charted the development of delusional thinking in a 72-year-old woman who became convinced the empathic-response bot was in love with her. The rise of “AI psychosis” has the potential to exacerbate mental health problems, particularly in socially isolated people, which a quarter of Americans over the age of 65 are.

ELARA’s research is focused on creating “pet-like” AI models with enhanced trust cues. If it can overcome the dangers of socially isolated people relying on AI for companionship, it could be a big step forward for independent aging.

SpaceX IPO set to be biggest ever and could make Elon Musk a trillionaire

IPO News

SpaceX says it plans to raise up to $75 billion when it goes public this month, setting the stage for the largest-ever stock market debut and putting Elon Musk on course to becoming the world's first trillionaire.

The company, formally known as Space Exploration Technologies Corp., said Wednesday it will sell 555.6 million shares at $135 a piece in an initial public offering. The estimated proceeds would easily top the $26 billion raised by oil giant Saudi Aramco in 2019. The offering would also give SpaceX a market value of $1.77 trillion. Only six companies in the S&P 500 are currently worth more, with Nvidia tops at $5.2 trillion.

Besides the size of the offering and the expected proceeds, SpaceX's amended prospectus updates details about how much control of the company Musk will have. As SpaceX's CEO, chief technical officer and chairman, Musk's voting power will come primarily through his ownership of 5.22 billion Class B shares, which give the holder 10 votes for every share held. According to the filing, Musk would have 82.4% of the voting power in the company.

Forbes currently values Musk's net worth at $826 billion and his stake in SpaceX at $542 billion. The estimated value of his SpaceX holdings was based on an overall value for the company of $1.25 trillion. Based on those numbers, a $1.77 trillion valuation for SpaceX would boost Musk's net worth by $223 billion, making him a trillionaire. However, much of Musk's worth is in stock that he has yet to cash in.

Even as it makes a bid for a blockbuster market debut, SpaceX is currently losing billions of dollars a year. The filing shows that the company lost $2.6 billion from operations last year on $18.7 billion in revenue, and the losses kept piling up at the start of this year, too.

Fantastical plans

Time will tell how SpaceX fares on the market. Musk's plans for the company are as fantastical as the money he hopes raise in the sale.

Colorful, even frightening in parts, the IPO document strikes a contrast with the typically dry, technical prose in IPO documents, detailing plans to use proceeds from the sale to help put men on the moon again and perhaps even Mars. In one section, it talks of a need to build "a permanent human colony" on the red planet with "at least one million inhabitants" as existential threats loom that could consign man to "the same fate as the dinosaurs."

Musk has almost equally ambitious plans for his other publicly traded company, Tesla. His goal is to transform the maker of electric vehicles into a producer of robotaxis and humanoid robots. Dan Ives of Wedbush Securities wrote in a research note that he expects Tesla and SpaceX to merge next year.

AI plays a key role

Key to the success of both companies — and any merged entity — is artificial intelligence. In its IPO filing, SpaceX says it sees potential revenue from AI of up to $26.5 trillion. But that depends on another lofty Musk ambition — putting data centers in space, which is not technologically possible at the moment.

Transforming his space company into a primarily AI-focused company will be a challenge for Musk, who started xAI in 2023 with 11 other co-founders who have all since left. Some were recruited away by rivals.

Its main AI product, the chatbot Grok, is "less impressive than anything that we see from any other major player in the space, whether that's OpenAI, or Anthropic, or (Google's) Gemini," said IDC analyst Arnal Dayaratna.

Dayaratna said that doesn't mean SpaceX doesn't have potential as a major AI player, thanks in part to its computing partnership with Anthropic and Musk's recent deal that gave SpaceX the rights to buy AI coding tool Cursor for $60 billion later this year. Folding in Cursor's capabilities would give SpaceX access to the coveted business customers now using Anthropic's Claude or OpenAI's ChatGPT.

SpaceX plans to use the net proceeds from the IPO to fund the expansion of infrastructure for its AI and rocket businesses, and to beef up the constellation of satellites that power Starlink Mobile, among other investments.

The company plans to list on the Nasdaq under the symbol "SPCX" and could begin trading as soon as the end of next week.

And SpaceX isn't the only colossal market debut investors are now bracing for. Earlier this week, Anthropic submitted a confidential filing with the U.S. Securities and Exchange Commission to officially start its own IPO clock.

OpenAI has not yet reported filing the initial SEC paperwork, but an IPO from the ChatGPT maker is widely expected.

"This listing represents the first major test for public markets after years of muted IPO activity with SpaceX paving the way for AI giants Anthropic and OpenAI to follow soon after," Ives wrote.

___

Associated Press Technology Writer Matt O'Brien contributed.

New UH survey reveals concerns over AI data center growth in Houston

data findings

A new report out of the University of Houston shows that area residents remain wary of the long-term effects of operating data centers.

The recent survey from the University of Houston’s latest SPACE City Panel, conducted by the Center for Public Policy at the Hobby School of Public Affairs, shows that while 85 percent of Houston-area residents use AI, nearly 63 percent oppose the construction of AI data centers within 1 mile of their homes.

Respondents’ concerns centered around data centers’ high energy demand and the area’s power grid reliability. According to the survey, 32 percent of residents who oppose local data center projects would be more likely to support the centers if they relied on renewable energy over fossil fuels.

“Respondents understand that AI can bring economic and educational benefits, but they are also concerned about the physical infrastructure needed to fuel AI, especially data centers,” Soran Mohtadi, post-doctoral fellow at the Hobby School and a researcher on the report, said in a news release. “This physical infrastructure demands more electricity and water, leading to environmental impacts.”

Experts estimate that 6.5 gigawatts of data center capacity will be added to the Texas grid by 2030. And Houston’s data center capacity is predicted to more than double by 2028.

The Electric Reliability Council of Texas also projects electricity demand could reach 218 gigawatts by 2031, which would be more than double the record peak set in August 2023. Data centers are expected to account for 86 gigawatts of that new demand.

Survey respondents also said they are concerned about the state's future water supply, given the large amounts of water that data centers need to stay cool.

In terms of who’s responsible for that issue, 57.6 percent of respondents said they put the onus on Texas lawmakers, while 31.5 percent say tech companies should be responsible.

Additionally, more than 75 percent of respondents believed that data center developers and technology companies—not residents—should bear the cost of infrastructure upgrades to support data centers.

“Every decision legislators make has implications on residents’ everyday lives and local infrastructure now and in the future,” Maria P. Perez Arguelles, lead researcher on the report and research assistant professor at the Hobby School, added in the news release. “This issue is going to become more important in years to come, so this is just the beginning.”

Read the full report here.

View All Listings