[lrae]

> Hasn't DeepSeek's novel training methodology changed all that? If the energy and financial cost for training a model really has drastically dropped, then frequent retraining including new data should become the norm.

Even if training gets way cheaper or even if it stays as expensive but more money gets thrown at it, you'll still run into the issue of having no/less data to train on?

[photochemsyn]

True. One effective test for AGI might be the ability to first create a new language, then also write performant code in that language.