Today I was reading about a new type of application for generative image models - reading your mind. See, the models are essentially decoders from noise - they decode from a specific latent space(or origin) to produce images, with that latent space being shaped by what the purpose is.
Now what if that space from which it decodes is your brain waves, and the training is images being shown to a person?
The answer is MinD-Vis. Developed by a team of scientists at NUS, CUHK and Stanford, this model can decode your brainwaves into what image you are looking at, at that point, by pretty much using the same technology as Stable Diffusion.
This is genuinely good - the images being reconstructed aren’t perfect, but then, have you ever tried to recall a memory? It looks kinda about the same in your mind. This technology is absurd, and it can’t be that long before the other generative models use this technology - imagine literally having anything you imagine being shown to you immediately.
Beyond text, there’s also a team at Caltech who developed a brain wave to speech model using pretty similar principles. So beyond Midjourney, we could see CHatGPT-3 getting so good that you can write a book by thinking about it. It’ll finally breaching the brain reality barrier! No more typing necessary.
Beyond the mind reading, there’s also GATO ,a model developed by Meta which breaks out of the siloed paradigm of AI modeling. To quote their article, “The same network with the same weights can play Atari, caption images, chat, stack blocks with a real robot arm and much more, deciding based on its context whether to output text, joint torques, button presses, or other tokens.”
This general of an application shows that general AI is already here and we’re just starting to see the potential. The training isn’t even a big issue - what human can do anything without being somewhat trained? - but the power is starting to get there. In 2084, we might be driving around in mind controlled AI cars overseeing vast assembly lines run and operated entirely by AI, listening to AI music, while watching AI movies. It’ll be an interesting new future.