Just had this failure with NanoBanana this morning. Asked for transparency and it drew the transparent image checkerboard. Didn't think to ask it to fill the background with a key color so GIMP et. al. could do the transparency. Just tried it out and it ended better than the first attempt.

Toyota wouldn't dare adopt flying cars (yet) but it's not hard to draw a line between Toyota and Joby Aviation.

No doubt. There is a vacuum around fast and Montezuma-class exploration challenge. Not sure if there's anything in the "ocean" that fits that. But no doubt that faster environment is a big boon.

Toyota has fancy research arms under a broad banner. Toyota is a mobility company and this their R&D. Very forward looking things are unlikely to fall directly under the Toyota brand unless it is of Toyota-level maturity. Might see it out of a brand that Toyota has heavy investment in, though.

Aside from the streams, is there a compelling reason why Pufferlib is doing a better job educating over Gymnasium? The "ocean" isn't very deep and feels disjointed. There's plenty to dislike about ALE but there's a breadth of problems there.

I only thank single-gpu models. It's endearing when they are effective.

The fact I still think Crysis is a relevant metric might date me.

I selfishly need this to work because I love small, cheap trucks. The more popular this is, the more likely an incumbent automaker will make their version of it. Electric Tacoma in this size would slap. I don’t need to remind Toyota they already make Kei trucks overseas.

Maybe all of the mental gymnastics it has to do to justify it’s positions is part of its apparent stellar training.

The core examples given in the article contain words such as “hypothetical” and “theoretical”. As a peer reviewer of this research I’d want to somehow verify that this isn’t just a replication of the well known “for a novel” hack.