>>22995007
>these bigass models the companies are using are dubious because of the model size and too much training data, you get overfitting
Interesting. So if you were to train a LLM on a math textbook, you'd get better results than if you trained it on the entire internet where it has the chance of taking in bad info?
>>22995052
>the captcha is kind of annoying and inconvenient, but it's not that bad, and you only have to do it once a day
See, I all my browsers set to delete cookies upon closing so I had to do them a bunch. Maybe that's why I stopped using it. Yandex had a few too but there wasn't a set pattern to it.