2 ms·
LLMs are still computer programs btw. There's the program that scrapes, the program that trains, the program that does the inference on the input tokens. So it
by TZubiri 2y ago
LLMs are still computer programs btw.
There's the program that scrapes, the program that trains, the program that does the inference on the input tokens. So it's hard to say exactly which part is responsible for which output, but it's still a computer program.
- fragmede 2y agoSimplistically, the output from the program that scrapes is the dataset, the output from the training program is the model, and the output from the combination of the program that does inference using the model is the LLM output - be it text or a picture or some other output (eg numbers representing true/false in a fraud or anti-bot or spam for a given input). ML models are relatively old, so that's not at all a new paradigm. Even the Attention Is All You Need paper is seven years old.