The math involved in LLMs is not complex for anyone that has passed undergrad Calc and Linear Algebra classes. If you know derivatives, the chain rule and some matrix basics you can figure them out with enough studying.
The hard part about LLMs is not the math but the neural net architecture innovations they brought (eg self-attention)
It can if you don’t do a train-test split.
But even if you consider the training set only, having zero loss is definitely a bad sign.
Not excusing Vanguard, but if you’re running Windows then your entire kernel is a blob. If you’re running most linux distros, then your kernel contains blobs for drivers.
not on my machine! every time someone posts a screenshot with a handwritten font it’s less readable and looks bad
Ian also happens to be dead
Not at all. The only similarity is that LLMs work with text, and the document formats can also represent text.
Each format (E.g pdf, json, excel) has a defined standard, so all you have to do to change between each other is to map one format’s fields to the others. You don’t need (and won’t get good results) from having an LLM produce the new format from scratch.
What he’s asking is the equivalent of asking if there’s an LLM made specifically for solving arithmetic problems. Why would you try to solve addition using an LLM?