

4·
12 小时前Not even that, it’s an approximation of the next word or words.


Not even that, it’s an approximation of the next word or words.


No AI is a made up term for statistic models tech Bros used to “train”.
There’s machine learning and a subsidiary of ML are LLMs, same as LDA, Random Forest, Decision Trees and other stochastic models.
The concept of those models isnt even remotely new, the first neural models were hypothesized and trained in the 60’s. Although with only a fraction of the layers of today’s models, since compute power wasn’t remotely close to what it is today.
These models only can exist because we (humanity) got so good at building tiny transistors.
And as a studied Data Scientist it boils my blood whenever some problem is supposed to be solved with “let’s use ai for it”.
And then, came BERT.
I’d love to be back at BERT levels of transformer networks.