This would be much more useful if it wasn't stickmen but animating components (which would allow to do "launch video" tier animations) on boxes, and websites. This would allow for people to create tiny components and not put videos. I feel like stickmen are a "nice to have".
OpenAI is going to close the one thing it needs to be profitable : calculation power. Love this website : https://isaiprofitable.com/, shows who wins at the AI revolution. Nvidia wins because it has instant revenue, OpenAI is going to close that gap.
I personally do this, ask claude code to teach me about concepts I don't know about when it codes something, and only then I accept what it suggests to me
The problem with eval is the fact that the information is not updating itself fast enough so that you want the latest model performance benchmarks. Bloomberg succeeded because it sells info that is expires in the next hour.
I like the part where they talk about the color perception space. The color perception space has itself multiple representations, according to the distance definition you are taking. The one shown in the video is the CIEXYZ (1931). Since human perception is non linear, there are other definitions which take into account those non linearities to make the space more "Hilbertian", such as CIELAB (https://en.wikipedia.org/wiki/CIELAB_color_space), or CIELUV (1976, https://en.wikipedia.org/wiki/CIELUV). This article sums up well the different representations : https://math.stackexchange.com/questions/1777699/color-space...
Interesting how they voxelate the splat to print it. I am sure there are more efficient algorithms that allow you to make a mesh which then gets sliced before printing
Fun fact for the first apple keyboard layout on the first iphone, the touchscreen hadn't the resolution to tell appart which letter you meant to type in, so it changed dynamically the "hitboxes" of the letter buttons when you typed a certain letter. (for instance if you typed the letter "i", the hitbox of "t", and "n" were changed to be bigger, because there is a high probability you were hitting those next. Here is an article that talks about it : https://www.cnbc.com/2017/06/26/creation-of-the-first-iphone...
one million Claude Tokens (assuming you are on opus) = 5 USD = the very dongle you tried to replace.
Add the cost of the rasberry pico, you'll have an easier time buying the wifi dongle. The project is cool thought to learn about networks, NAT, Proxys, ect...
Now is precisely not the time to cut jobs, but to invest in people's usage of AI. You have domain expertise, and you are going to burn tokens trying to replace people's jobs, but the AI revolution is not at all about people replacement, its about a change of paradigm. Watch them closely rehire thousands of new employees (who won't have the domain expertise) once they see that you actually need people to operate their company :)
I find that this kind of web design is not conforting, it brings back memories of slow loading pages, scammy websites that download things without warning you.
This feels like a classic “everyone becomes a cloud provider” extrapolation that ignores how little of cloud is actually about owning compute and how much is about operating a full dev platform.
Apple isn’t going to turn itself into AWS just because it has servers. This would require a complete shift in their business model, and I highly doubt their will to run a messy infra business it has never shown interest in. SpaceX and Apple have completely opposite cultures.
10 new reactos by 2040 is huge when you come to think of it, this is just in 14 years, this is about 0.72 reactors built each year. Knowing you need to build : the whole reactor building, the machines building, the whole plumbery system, the electrical system, the control command system, the reactor, reliability features, and all of that knowing you have to pass the nuclear regulation authorities controls.
The way the problem was solved at first hand by just "recognizing the pattern of (2n) choose n" wouldn't satisfy me at all, where's the proof ? Why does it work ? This isn't maths, this is "pattern recognition".
This is very cool, it would be huge if this camera could go on fields "traditional cameras" didn't go in, and iphones cannot give. Like an inbetween both where you can add an objective, and send photos to your phone via BLE or something. This would allow for a "professional camera" where you can edit your photos you just took
What I find cool is not the model in itself, but the architectures / training methods found that make the model better. It gives out a new possibilites for other fields of AI. (Notably if you want to fine tune other CV models)
Frankly after testing out Fable last week, it was just a bigger sink of tokens than anything else. The amount of tokens consumed by it wasn't worth the steps it saved me compared to using opus 4.8.