This week - Nvidia reveals a new GPU for AI; worms detect lung cancer; a crazy robot from Switzerland; deep learning is hitting a wall; and more!
More Than A Human
Thanks to a brain implant, a paralysed man was able to communicate again with his family and the world in general. By thinking about moving his hand, the patient was manipulating his brain waves which then were picked by the sensors and translated into simple yes/no answers to groups of letters, and then individual letters. After about 3 weeks with the system, he produced an intelligible sentence: a request for caregivers to reposition him. In the year that followed, he made dozens of sentences at a painstaking rate of about one character per minute: “Goulash soup and sweet pea soup.” “I would like to listen to the album by Tool loud.” “I love my cool son.”
Stanford released the 2022 AI Index report. The report itself is 230 pages and here are some key takeaways: private investment in AI soared while investment concentration intensified; U.S. and China dominated cross-country collaborations on AI; language models are more capable than ever, but also more biased; AI becomes more affordable and higher performing; more global legislation on AI than ever.
Nvidia has announced a slew of AI-focused enterprise products at its annual GTC conference. The main star of the event was Nvidia's new silicon architecture, Hopper, which compared to the current Ampere lineup of chips presents a massive jump in raw power (AnandTech has a nice breakdown of H100). Nvidia also teased a new supercomputer, named Eos, which will be built using the Hopper architecture and contain some 4,600 H100 GPUs to offer 18.4 exaflops of “AI performance.” The system will be used for Nvidia’s internal research only, and the company said it would be online in a few months’ time.
In this article, Gary Marcus argues that deep learning has reached a wall and we need new approaches to move the field forward. His solution is to look back again at the idea of symbolic AI - systems that combine the power of deep learning with symbol manipulation approaches.
Researchers from ETH Zurich present their new quadruped robot that instead of feet has wheels. It can ride on all four wheels and, when it is needed, it can "stand" on two wheels and use the other two wheels as "hands". Very interesting robot.
By taking advantage of nematodes' extraordinary sense of smell, researchers were able to build a device with worms inside of it to detect lung cancer cells. This "worm-on-a-chip" could someday help doctors noninvasively diagnose cancer at an earlier stage.