I’m no fan of meta, but a reminder that they are one of the best right now for keeping their AI developments more open and available. This is thanks to yann lecun and other researchers pressuring meta to keep their info on the subject more open.
Are we looking to punish them for making their work accessible?
Not to mention how important something like joint embedded predictive architecture could be for the future of alignment and real world training/learning. Maybe go after other foundation model developers to be more open, if we’re complaining about the inevitably public nature of some information within the mountainous datasets being used.
Although I’m still of the mindset that the model intent matters more than the use of openly available data in training. I.E. I’ve been shouting about models being used specifically to predict and manipulate user interactions/habits for the better part of a decade. For your “customized advertisements” and the like.
The general public and media interaction on the topic this past year has been insufferably out of touch.