That's not exactly what Lecun said, he simply was criticizing the commercial approach of Meta and the lack of research. He himself thinks he can do more with just a different approach, not "unknown" tech.

Meta, however, is only one of the companies working on it and they seem considerably behind competition currently.

People complaining that things crash is in fact very anecdotal and, even if it was true, LLMs being not that good yet isn't news and we aren't talking about that.

Same goes for the fact that LLMs aren't anywhere close to AGI, like marketing teams try to pretend.

On the other hand though, I have heard university professors saying that coding through LLMs is improving constantly and so rapidly that they think it is very likely that they will become widely used tools even for professional programmers in the near future.

The time saving is game changing for any such activity.

Should be also noted that we are seeing big improvements in time frames of months and LLMs have been "mainstream" for not that long yet. It's so early to say what can or cannot be achieved in my opinion.

Lastly, I'm not saying the bubble won't pop. If I had to make a prediction the bubble will pop, some companies will burn to the ground but some others won't because they were playing a different game the whole time (Google is the first example that comes to mind), which is why I think it won't be the catharsis every anti-AI person is waiting for.

As for AGI, I would dare to say that we'll get there too, but it's a useless prediction with no value.