The thing that annoys me most is that there have been studies done on LLMs where, when trained on subsets of output, it produces increasingly noisier output.

Sources (unordered):

• What is model collapse?
• AI models collapse when trained on recursively generated data
• Large Language Models Suffer From Their Own Output: An Analysis of the Self-Consuming Training Loop
• Collapse of Self-trained Language Models

Whatever nonsense Muskrat is spewing, it is factually incorrect. He won’t be able to successfully retrain any model on generated content. At least, not an LLM if he wants a successful product. If anything, he will be producing a model that is heavily trained on censored datasets.

I’ve been trying to reach you about your car’s extended warranty

Yeah, JPEG uses a convolution to effectively average the values. I think there are other ways, though I cannot remember the context or file format, which use some interesting concepts from linear algebra. I recall a professor telling me about using singular value decomposition in the process. But that’s a different topic.

I didn’t know WEBP also supports lossy compression. That makes it even more flexible.

Not really.

JPEG is a lossy compression format. It loses information to reduce filesize.

PNG is a lossless compressed format. It serves as a well-rounded format for general purpose image compression without the loss of information. The downside is the image can be much large in file size.

SVG is a vector graphic, as you seem to be aware. These files have what is effectively infinite resolution, but can be significantly larger in filesize, depending on the circumstances.

WEBP is a more efficient lossless compression format. It is analogous to PNG, but smaller file size. Additionally, as you stated, it can also be used for animated graphics,like GIF formats.

I think WEBP is a decent format and is significantly more modern than PNG. That being said, however, the main issue is the lack of modern integration and adaptation for newer image formats.

Personally, I use whatever. The only exception is when 8 need to store images on github in a repository. Then, I will typically convert to webp and optimise the image to reduce the file size as much as possible.