I would think there would be some interaction between the carburetor air intake and the propeller right next to it. Intentional feedback, maybe, to change the mixture as the RPM goes up? If not intentional, then it sounds like a potential showstopper.
It will generate anything. Xi/Pooh porn, Taylor Swift getting squashed by a tank at Tiananmen Square, whatever, no censorship at all.
With simplistic prompts, you quickly conclude that the small model size is the only limitation. Once you realize how good it is with detailed prompts, though, you find that you can get a lot more diversity out of it than you initially thought you could.
Absolute game-changer of a model IMO. It is competitive with Nano Banana Pro in some respects, and that's saying something.
I could imagine the Chinese government is not terribly interested in enforcing its censorship laws when this would conflict with boosting Chinese AI. Overregulation can be a significant inhibitor to innovation and competitiveness, as we often see in Europe.
One really-cool way to solve that problem is to embed a 7-segment LED under each keycap. You walk up to the keypad and the 0-9 digits appear in random order. No one can shoulder-surf, look for wear or IR emission from the buttons, or train on the click sounds.
Dell had those on every lab door in the building back in the early 90s. You felt like 007 every time you punched in your access code. I've never seen them anywhere since.
And now days I can't put in my card's pin without 10 overhead cameras aimed at the register area. All the cameras of which are network-connected, video stored persistently, and high res/fidelity enough to here the little beeps as I press the keys, and to know that I've hit the enter because the screen indicates it immediately. But then Dell cared about its own security, and the grocery store doesn't give a single shit about whether my life is ruined by identity theft.
Not necessarily. The problem may be as simple as the fact that LLMs do not see "dog legs" as objects independent of the dogs they're attached to.
The systems already absorb much more complex hierarchical relationships during training, just not that particular hierarchy. The notion that everything is made up of smaller components is among the most primitive in human philosophy, and is certainly generalizable by LLMs. It just may not be sufficiently motivated by the current pretraining and RL regimens.
They have had to walk that back, because it cost them dearly in market share. Turns out most of their customers don't agree that touchscreens are unwanted.
There's also the fact that the problem was never about touchscreens per se, but inappropriate/incompetent UI design that happened to use touchscreens.
reply