My take on the news:
Meta will introduce AI watermarks: They have already developed watermarks for AI-generated images that are not visible to the human eye, and now they are about to deploy the watermarks on Facebook, Instagram, and Threads. Most big tech companies have proposed or joined standards for AI watermarks, but I think 2024 will become the year when this initiative is scaled to the public. AI images are already polluting the internet, creating a big problem: AI image generation is fed by internet images, but what if those images are themselves AI-generated? Then, the disinformation problem will be big during the 2024 elections in the US, Mexico, and other countries.
Anthropic’s Claude 3 appears to outperform its rivals but can't find its own subscription plans: Claude's third iteration is out (in case you haven’t tried it, head to claude.ai), and according to reports, it outperforms GPT-4 and Gemini at some tasks. But I asked Claude about the difference between the free and the paying plan, and it told me “I'm afraid I don't actually have information about different subscription levels or pricing plans.” Then I asked Google’s Gemini the same question about Claude, and Gemini gave me the right information. How is this possible?
MagicLab robot can fold clothes and more: MagicLab is just one of several humanoid robots that can perform domestic chores such as folding clothes. Of course, this is a technological feat, but just how important is it to fold clothes? And how important is it to be humanoid? The clothes thing is not critical per se, but it showcases the capability of handling delicate and non-uniform objects like blouses or undies. The humanoid aspect is taken by many as “cute,” but, more importantly, it leverages the fact that most objects in humans’ lives are made for humans’ proportions, like stairs, doors, faucets, cars, you name it. For instance, you can build a robot 20cm tall, but it won’t be able to make a bed.
Quote of the week
If you have read my previous posts, you know for a fact that I’m not a fan of Elon Musk, mostly because of his iterative false promises. But I stumbled upon a very good quote from him:
“Prototypes are easy, volume production is hard, positive cash flow is excruciating,”
– Elon Musk
You can tell this to anyone who comes with a prototype as proof of a viable business…
This week’s product
The idea of asking a PDF about its content is not new, and several prototypes were announced in 2023. But now, Adobe has launched an extension for Acrobat that allows you to ask questions about your PDFs. As I have commented before, there are two main ways of launching AI products: one is a new startup devoted to the product, and the other is a big company giving additional functionality to their products, and this is the case here.
You can access the product (now in beta) here, but you’ll have to give them your credit card information so you will become a client after a free trial at the rate of 20 dollars per month.
What is…?
HoloTile.
HoloTile is a real innovation developed by… (you won’t believe this) Disney!
Yep, and it’s not an imaginary object like the laser sabers of Star Wars, but a real-life treadmill floor that can move around people or objects.
HoloTile solves the following problem: when you put on a VR headset, you won’t see the real world, so it becomes hard, even dangerous, to stand up and move.
Not anymore with the HoloTile floor.
It is a multi-person, omnidirectional, modular, expandable treadmill floor. It’s “composed of hundreds of independent, motorized tiles that can shift and rotate beneath your feet, seamlessly keeping you in place,” according to Disney.
You can check a nice video of it on this link.
This week’s blog piece
Yesterday, I published the article “Will Ever Self-Driving Get Back its Shine?” on Medium. Its main highlights are:
Apple’s car “secret” project, Titan, shut down shortly after Cruise, the autonomous car GM division, closed for good amid a scandal about secret remote operators for its cars…
It’s clear that self-driving initiatives are facing trouble and also that billion-dollar investments are going nowhere.
Tesla also had its share of accidents.
Why Machine Learning approaches to self-driving are having trouble with “edge cases.”
Edge cases are supposed to be rare, but they happen constantly.
There is a human bias where we confound randomness with uniformity, so we don’t consider edge cases.
In the end, a few companies related to self-driving will survive, but once they are alone in that space, their position will be stronger than ever.
Tesla and Waymo could be the last survivors. The article includes a Larry Page anecdote I recounted in the newsletter.
Here is the end of the free preview of SkepTech. Please become a paying subscriber for more exclusive content, including a “Friend Link” to the mentioned article and links for some curated articles on Medium.