This is the high-level explanation of the simplest diffusion architecture. The model trains by taking an image and iteratively adding noise to the image until there is only noise. Then they take that sequence of noisier and noisier images and they reverse it. The result is that they start with only noise, and they predict the removal of noise at step until they get to the final step (which should be the original image (or training input)).
That process means they may require a hundred or more training iterations on a single image. I haven't digested the paper, but it sounds like they are proposing something conceptually similar to skip layers (but significantly more involved).
I'll link to an open source project [0] because I don't want to throw mud at any particular company, and it is a cool project to develop for anyone motivated enough.
Having said that, it's a dumb product because the cane is very often beaten up, or misplaced so it's an expectation that it will need replacing every 12-18 months. The additional weight needed for the hardware throws off the natural balance of the cane making it more difficult to use and also interferes with the haptic feedback users depend on from a "dumb" cane. Also, the white cane is relatively cheap ($20-50 USD) - smart canes are usually $1000+, or if the "smarts" come as attachments to an analog cane, they would range $300-500 (my intelligence is 2-3 years old, so it might be higher now), so out of reach to most blind people. None are stand-alone and require synching with the phone, but many users who are tech savvy enough to use one of these already uses a navigation app, so one of the key benefits (GPS navigation) is redundant.
Overall the drawbacks (price, negative impact on usability, natural wear and tear) greatly outweigh the possible benefits.
Everyone in the industry is genuine, there is very little profiteering, and the goal across the board is to improve people's lives, but these products are always from young techies (student or technically advanced blind person) who made an awesome project that should be celebrated. But they should also be one-offs and open-sourced, with the money and effort spent on launching smart canes that will inevitably fail in the market better spent elsewhere.
Super Intelligence would expand at near the speed of light, reaching us extremely fast. It would be unlikely for the window of time that we have telescopes and the window that the aliens are expanding towards us to overlap.
Intelligence is consistent with the laws of physics but given our rudimentary understanding of it we have no idea if super-intelligence is. I'm defining super-intelligence as at least an order of magnitude increase in intelligence, not just "high IQ".
Intelligence is correlated with brain size, not just in humans, but also between species. No physical law prevents us from building the equivalent to a house sized brain. Perhaps not even a moon-sized brain. The only real limit seems to be mass (we don't want the whole thing collapse under its own weight) and signal delay due to the speed of light.
> No physical law prevents us from building the equivalent to a house sized brain
I think you'll find the speed of thought is much lower than the speed of light, I believe nerve impulses max out at about 120 m/s. The difficulty of supporting such a large mass of nerve tissue is also a problem. There is also the trouble of whether intelligence is purely electrically mediated or if it relies on the soup of hormones and molecular machinery of DNA and proteins for some of its functions. We also need to specifically consider the neocortex here since it is the only differentiating brain structure in animals with human intelligence. Other animals have much larger brains but no signs of human like abilities.
If you are suggesting we build an artificial brain then we need to wait at least a few more years if not decades to find out if that is possible. The current transformer models are quite advanced at language production and seem to have some simple reasoning abilities but they are very far from being intelligent.
I mean, we don't need to call our house or Moon sized AI "brain", but it seems clear that no laws of physics are preventing us from building something so intelligent that we are mere ants in comparison. The old Gods, while powerful, would pale in the light of its intelligence. A true God, hopefully a benevolent one. It doesn't seem many decades away.
Different. Super-intelligence is a plausibly reachable thing, a question of engineering. TIL is an uncertain thing, requires entirely new physics, and completely unknown tech.
reply