Fabio Manganiello on Nostr: In today’s edition of “just because you know how to use a hammer it doesn’t ...
In today’s edition of “just because you know how to use a hammer it doesn’t mean that the whole world is made of nails”: LLMs may actually add no value to time series forecasting than more traditional techniques (like multivariate regression, or even a simple attention layer without a full blown LLM downstream). And in many cases their biases may actually make things worse.
Just because a #LLM is good at predicting the next token in natural language contexts, it doesn’t necessarily mean that it’s good at predicting the next numerical value in a time series.
https://arxiv.org/abs/2406.16964
Just because a #LLM is good at predicting the next token in natural language contexts, it doesn’t necessarily mean that it’s good at predicting the next numerical value in a time series.
https://arxiv.org/abs/2406.16964