> It’s hard to collect enough data using invasive methods. Few people want a chip in their brain! But non-invasive methods are getting much better. The hardware is improving and getting cheaper, though I apologize for being vague about the particularities of our hardware.
Their data acquisition routine was driven by Claude and they were recruiting random people off Craigslist pretty aggressively a few months back. I know a couple of folks who responded out of curiosity and described a very janky setup in the rented basements of a couple of residential houses in SF. From what I heard about the setup the data they got must have been incredibly noisy.
To the people who are sceptical here, I wouldn't count them out yet. They almost certainly can get funding with what they captured, which would get them better hardware. And given the state of brain scanning - people have captured images from scanned brain activity some years ago as I recall - I'd say they have a good shot. But yeah, it's not telepathy.
how do you pick good goal conditioning images/do you have to hand pick a dataset of good goal images? seems hard if you don't have full context. really cool though!
you don't need to pick good images manually! it's similar to next-token prediction, many prediction tasks aren't especially interesting, but there are enough hard ones that the model can spend the most time learning from those. the simplest thing scaled up works very well.
we have an alignment blog post dropping soon! scaling up in the next couple of months, then hopefully opening up an API or licensing it.
Benchmarks are really fun—lots of secret ones. Our main thesis is that you should be using the same benchmarks to measure human ability to use a computer, as you would an AI model. Definitely a suite of continuous long term planning tasks (games) and things such as marking emails as spam etc.
definitely! we are looking into more interp + visualizations in general as we scale up.
Frankly, this comment does little to avail the parent's point - doing this on the open road was both illegal and reckless. It reflects extremely poorly on your character and the project as a whole.
thanks! the math and architecture of the FDM (no video encoder) is pretty simple, its a regular transformer with next-token predictions but with frames interleaved.
