Main Highlights
- ThirdAI earned $6 million in seed investment for its tools for speeding up deep learning technologies without the requirement for specialist hardware such as graphics processing units
- The investment will be used to hire new workers and invest in computer resources, according to Anshumali Shrivastava, co-founder, and CEO of Third AI
- ThirdAI accelerator creates hash-based processing methods for neural network training and inference
- ThirdAI’s algorithmic breakthrough has shown that they can make commodity x86 CPUs 15x or quicker than the most powerful NVIDIA GPUs for training massive neural networks
ThirdAI, located in Houston, earned $6 million in seed investment for its tools for speeding up deep learning technologies without the requirement for specialist hardware such as graphics processing units.
The investment was co-led by Neotribe Ventures, Cervin Ventures, and Firebolt Ventures, and will be used to hire new workers and invest in computer resources, according to Anshumali Shrivastava, co-founder, and CEO of ThirdAI.
Shrivastava, who has a background in mathematics, has long been interested in artificial intelligence and machine learning, particularly in considering how AI may be created more efficiently. He investigated how to make that work for deep learning while at Rice University. He founded ThirdAI in April with the help of three Rice graduate students.
Third According to Shrivastava, AI technology is intended to be a “smarter approach to deep learning,” combining algorithm and software improvements to make general-purpose central processing units (CPU) quicker than graphics processing units for training massive neural networks.
Companies abandoned CPUs years ago in favor of graphics processing units, which could generate high-resolution pictures and video simultaneously more quickly. The drawback is that graphics processing units don’t have a lot of memory, and customers frequently run into bottlenecks while attempting to build AI, he noted.
“When we looked at the deep learning landscape, we saw that much of the technology was from the 1980s, and the majority of the market, about 80%, was using graphics processing units, but we’re investing in expensive hardware and expensive engineers and then waiting for the magic of AI to happen,” he said.
ThirdAI’s Algorithmic Breakthrough
ThirdAI is a cutting-edge artificial intelligence firm focused on scalable and long-term AI. ThirdAI accelerator creates hash-based processing methods for neural network training and inference. The technique is the product of ten years of research and development into efficient (beyond tensor) mathematics for deep learning.
ThirdAI’s algorithmic breakthrough has shown that they can make commodity x86 CPUs 15x or quicker than the most powerful NVIDIA GPUs for training massive neural networks. The discovery has challenged the AI community’s long-held belief that specialist processors such as GPUs are far better than CPUs for training neural networks.
Third AI innovation would not only assist present AI training by migrating to lower-cost CPUs, but it should also allow the “unlocking” of hitherto unfeasible AI training workloads on GPUs.
He and his team studied how artificial intelligence (AI) was expected to evolve in the future and sought to build a low-cost alternative to graphics processing units. Instead, their technique, dubbed the “sub-linear deep learning engine,” makes advantage of CPUs that do not require specialist acceleration gear.
Neotribe’s creator and managing partner, Swaroop “Kittu” Kolluri, stated that this sort of technology is still in its early stages. Current techniques are arduous, expensive, and sluggish, and if a firm runs language models that demand more memory, for example, it will run into issues, he noted.
How does ThirdAI help you?
“That’s where ThirdAI comes in because you can have your cake and eat it,” Kolluri explained. “It is also the reason we decided to invest. It’s not just the processing, but also the memory, and ThirdAI will make it possible for everyone to accomplish it, which will be a game-changer. There is no limit to what is conceivable as deep learning technology becomes more sophisticated.”
AI is already at a point where it can handle some of the most difficult issues, such as those in healthcare and seismic processing, but he observes that there is also a concern regarding the climatic consequences of running AI models.
“Training deep learning models can cost more than five vehicles in a lifetime,” Shrivastava added. “As we move forward with AI scaling, we must consider those.”
