CES 2025 Day 1 – Peaceful progress

Generative AI is not yet ready for the consumer

Prototype to product

• The halls are busy and the queues are long but what is different in 2025 is the change in maturity of the products that are on display.
• Most of the time one sees prototypes stuck together with tape and chewing gum which work some of the time and have a tendency to feel like they are going to fall to bits when you pick them up.
• 2025 is different as many prototypes have given way to products that look well-built, work all of the time and are either already on sale or about to reach general availability.
• Most impressive is that the sprinkling of AI pixie dust that I predicted yesterday has not materialised with most companies only using AI for very practical purposes.
• A good example would be a series of pool-cleaning robots that use AI to map the layout of the pool to decide the best way to clean it or a blood pressure monitor that customises a standard dataset to give specific indicators of heart health beyond simple measurement.
• This is the kind of usage of AI that will make money now as opposed to massive inference to do PhD mathematics at very high cost.
• What this indicates is that generative AI services are not really ready for mass-market consumer adoption which is why they are not featuring as prominently as I had expected.
• This concurs with RFM Research’s conclusion that at the moment, the current use case for generative AI lies almost exclusively in the enterprise where the database and automotive are the ones that I think will feature first.
• This does not mean that generative AI won’t go to the consumer at scale eventually, but I think it will take time to penetrate leaving enterprise to make all the running in the meantime.

Robotics

• This plays directly into robotics which is a theme that is still in its infancy but with every CES that passes grows its presence and its relevance.
• This year there are more robots than ever on the stands albeit almost all of them are single-purpose products mostly in the household and garden cleaning category.
• Robotics faces two main challenges the first of which is locomotion which is only partly solved and the second is intelligence which I think is also, only partly solved.
• It is possible to teach robots to walk on legs without too much difficulty, but the problem is that every robot that is not absolutely identical has to be reprogrammed from scratch.
• This makes it very expensive to create a fleet of robots that use legs which I think is essential if robots are going to participate in the human world in a meaningful way.
• Some form of general framework is needed to generalise locomotion, which is what Nvidia Cosmos is all about.
• By creating realistic video feeds with data from a digital twin in Omniverse or other sources, the hope is that it becomes much cheaper to train robots and even make progress on generalising robotic locomotion.
• Nvidia believes that robotics is going to be a massive opportunity and in the very long run, I agree.
• However, there remain substantial hurdles to get there and by being one of the first to try and solve these problems, Nvidia is setting itself up to be the dominant supplier of silicon and software platforms for the robotic industry.
• The other issue is intelligence which I don’t think LLMs are going to solve on their own although they provide an excellent solution for a voice-based man-machine interface (MMI) through which robots can communicate with humans.
• RFM research indicates that the MMI problem has largely been solved (see here) but in terms of intelligence and behaviour, I am far from being willing to let one loose in my house.