joshwa
Joined 19 years ago
Joined 19 years ago
Individual models may have supplemented their training with things that look like structure (e.g. Claude with its XMLish delimiters), but it's far from universal.
Ultimately if we want better fidelity to the concepts we're referencing, we're better off working from the larger/richer dataset of token sequences in the training data--the total published written output of humanity.
I made my own version of this a while back, and it lets you create your own cutting methods, plot the statistical distribution, and share your ideas via permalink. It also lets you tweak onion parameters, such as number of layers and the layer thickness distribution curve).
Along the way I discovered two things:
1. I came up with my own method ("Josh’s method" in the app above) where the neither the longitudinal cuts nor the planar cuts are full depth, so the number of cuts at the narrower core is less than at the wider perimeter.
2. After all this hyper-optimization about size, it turns out what really matters when cooking is the THICKNESS, since ultimately determines the cooking rate. The only way to avoid thin outliers that burn long before the rest are cooked is to discard more of the tip of the onion, where the layers are the thinnest.
The 3D version of the simulator is still in progress--turns out 3D geometry is a lot harder than 2D. :)
Pull requests are welcome! https://github.com/joshwand/onion-simulator
I got a cheap MLX90640 off aliexpress for target detection and a grove vision AI V2 module to use with IR cam for classification/object tracking. Esp32 for fusion and servo/solenoid actuation.
Collab?
2. Write a new tool that is dead simple, opinionated (your opinions), convention over configuration, etc
3. Post to HN
4. Achieve adoption
5. As more people use the tool, feature creep ensues
6. In order to satisfy diverse use cases, make everything modular and configurable!
7. Tool slows and becomes impossible to manage
8. GOTO 1.
• Staff PM/Engineer
• [email protected] • http://joshwand1.substack.com/ • https://www.linkedin.com/in/joshwand