

That’s a bit harsh. A lot of the pictures you see on menus are not what they cooked in the kitchen. Because cooked food doesn’t look sexy enough. Or only when it’s hot and then steam coming off it makes it look worse in the image. So they use glue to gloss things up, replace entire ingredients (e.g. motor oil for maple syrup), etc. Food photography is serious business. And expensive.


This is about timelines. In the short-term it’s a win, in the medium-term it’s a threat, in the long-term it’s potentially catastrophic. They need money now.
This sort of scenario assumes that these companies will always be in a stronger position. I don’t think that’s a given. The companies that plundered all data everywhere, torrented what was available, and reluctantly bought what couldn’t be stolen to train their models are being passed on the right by companies that just look to steal or copy their stuff (e.g. China is inferencing the hell out of them). The hostility towards data centers also forces more development away from centralised servers onto local environments. At the same time, stuff is also being released as open source. I think there is a big chance that the big companies and their models today will be mentioned in the first chapter on the history of this technology but will fall by the wayside rather quickly. Their market position is underpinned by a belief that in the future they will be raking in the dough on subscription fees once they get the dirty work of being started here financed, no matter how. That’s a mighty big bet. Not only is a bubble economy a problem, there is no guarantee to protect their “IP.” It’s really like investing in very expensive tulip bulbs.
Also, we don’t fully understand this technology. A lot of the experts don’t either. Anybody who says they’re pretty sure they know where this road will take us in the next five years has ulterior motives.