The coolest — and arguably most useful — new technology at CES 2025

Nvidia lays out the future of AI

One of the most interesting technologies at the show was not difficult to find, because it was unveiled at the NVIDIA keynote. Company founder and CEO Jensen Huang did an amazing job showing off his vision for both the near and far future of artificial intelligence.

According to Huang, AI has a growing problem. Whenever anyone submits a question to an AI like ChatGPT or any of the other models available today, that query is reduced to tokens. Tokens are basically pieces of words, although they are cut up in odd ways depending on how the AI model works. When the AI generates a response, it does so in tokens as well, which are then assembled into words that the user can understand. The problem is that every token requires computing power, and millions of AI users are overloading data centers with too much token generation.

To compensate, NVIDIA unveiled the Grace Blackwell NVLink72 device. The golden-colored square is packed full of AI-optimized GPUs and weighs just over 1.5 tons. It’s also comprised of over 600,000 parts and two miles of copper cables. When powered up, it pulls in 120 kilowatts of electricity. It’s only job is to generate AI tokens, and the NVLink72s are currently being manufactured as quickly as possible at 45 factories around the world for delivery directly to data centers.

According to Huang, the NVLink72 devices are needed because generative AI is about to advance to the next stage of its evolution, what he calls Agentic AI. In Agentic AI, the main interface that humans interact with is only the top-level. Once a query is made, it is handed over to a bunch of sub-AIs which are each specialized in different areas. These sub-AIs, or agents, then run a bunch of simulations and permutations. They also talk to one another and share their findings while working on the core problem. And when those agents talk with each other, they do it at machine speed, meaning they can generate thousands of new tokens every second.

“When humans are using AI and tokens are generated at 20 or 30 tokens per second, that is okay because it’s as fast as you can read. But now that AIs are talking to themselves, token generation rates are going way up,” Huang said.

Huang predicted that in the business world, the agent-based AIs would eventually become almost like digital humans or employees, with each one assigned to different parts of a company to help with specialized tasks like cybersecurity or accounting. And in that future, IT departments will actually evolve to be more like human resources departments, only they will manage, onboard and train all of the various AI-based digital workers.

Assistive technology becomes real

I’ve always been a big supporter of assistive technology ever since I was a federal reporter covering Section 508 issues and devices. The CES show has not traditionally been a great place to find new assistive devices, and most of the ones that I have seen over the years have been prototypes that never actually became available for people to use.

That is what is so surprising about the GyroGlove. It was a prototype just last year at CES 2024. And now the glove is available for sale to people who are suffering with tremors or hand shaking caused by Parkinson's Disease and other neurological difficulties.

The glove is both simple in design and also packed with incredibly advanced technology. At its core, it’s a fabric glove that straps easily onto someone’s hand. A user’s fingers and thumb remain free and stick out from the glove. On the back of the glove is a gyroscope and a battery pack. How it works is that the gyroscope spins four times faster than a jet turbine, and works to counter the shaking movements or tremors that often occur when people have Parkinson's Disease. The battery life for the gloves lasts for about four hours on a single charge.

The company shared quite a few success stories, including people who could no longer do simple tasks like hold a cup of coffee because their hands shook so badly. But when equipped with the gloves, their hands were completely still once more. One woman with bad tremors was even able to return to her knitting hobby, something that she thought she would never again be able to enjoy. 

Robots charge though the uncanny valley

Last year the United Nations held a press conference where the stars of the show were robots, or more specifically, robots powered by artificial intelligence. The robots answered questions from human reporters, which were generally about how the robots really felt about humanity.

But as exciting as it was to hear from those robots, all of them were easily identified as artificial humans. Some wore dresses, hats and other fashion accessories, but their movements and their looks gave them away as not being real people. Arguably the most advanced robot on the platform that day in terms of AI, a robot named Ameca from Engineered Arts, was just a head sitting on a table. She had left her body behind for the press conference.

One interesting thing about the robots at the UN conference was that none of them got anywhere close to crossing the uncanny valley, a principle that states that, as objects become more humanlike, they risk putting people off with their creepiness. I’ve talked with robot developers before, and they have told me that they generally try to make sure their creations don’t look too human for just that reason.

At CES, one company called Realbotix aimed to blow right through that valley with a series of robots designed to look and even feel like real people. The newest model at CES was a tall, blonde, female robot named Aria. She was able to move around realistically, make hand gestures and generally interact with her environment. 

Aria has cameras mounted in her eyes, different from most humanoid robots where the camera is placed in their torso. Because of that, Aria can actually look at people while they are talking with her and react to them. If you smile while talking with her, she will generally smile back. And if you are doing something like eating an apple, Aria can recognize it and ask you how it tastes. Not surprisingly, Aria is designed for companionship, although she might also be employed in places like hospitals, nursing homes or even theme parks.

From the reaction of people packing in to see her and her companion, another female robot named Lyrica, it seems like the uncanny valley has been successfully crossed. While Lyrica and Aria’s movements seemed a bit stilted, they were also pretty human looking. Both are driven by a proprietary AI which does not seem that advanced compared with the robots at the UN conference. When questioned, both Aria and Lyrica had to pause to think about their responses for a very long time, and sometimes gave nonsensical answers when interviewed by reporters. However, they looked pretty when doing it, which gives a good preview of where at least one branch of robotics may be heading in the near future.

A bright personal pick

Those who have visited my home office or testing lab know that I love different lighting solutions. One of the happiest days I ever had as a reviewer was almost 15 years ago when I got ahold of one of the first Phillips Living Colors Generation 2 LED Lighting orbs, almost a full year before they went on the market. This was at the dawn of programmable LED colored lights, and when I saw those orbs at another trade show, the now defunct Comdex, I begged Phillips to let me have one. 

It took almost a year of nagging them, but they finally sent me one. I remember dancing around the lab with joy when it finally arrived. And it stood the test of time too, as it still lights up my office in brilliant blue hues, or whatever color I want, although I mostly use shades of blue these days.

Now that you know my love for lights, you will probably understand my personal pick for one of the coolest gadgets at CES. The Govee Gaming Pixel Light looks a lot like my old Lite-Brite toy from when I was a kid. Only instead of putting little plastic pegs on the screen to make images, the Govee light has large pixels right there on the screen. It comes with a few hundred images, and you can also upload your own GIFs. It can also run animations at 30 frames per second.

But the coolest thing about the Gaming Pixel Light is that it’s bundled with a pretty advanced form of generative AI. You can tell it what you want to see, and it will generate that picture for you in colorful pixel art. Sure, there is not too much practical use for such a device, but sometimes it’s good just to have technology that can make us happy. The Gaming Pixel Light is expected to go on sale for under $100 sometime in the next couple months, and will surely be installed in my office soon afterwards.

John Breeden II is an award-winning journalist and reviewer with over 20 years of experience covering technology. He is the CEO of the Tech Writers Bureau, a group that creates technological thought leadership content for organizations of all sizes. Twitter: @LabGuys