Hacker News Reader

Comment by nitwit005 | Hacker News Reader

Their data collection is naturally based off of people talking or typing: https://condu.it/thought/10k-hours

So you'd need to force people to somehow think the same things they do when talking or typing, without actually performing those actions. If you can do that, why not just have them talk?

View on HN

Isn't that just for data collection though? I imagine the fully trained model they're expecting to have will be able to map thoughts -> text. The data collection will naturally need to have the 2 sets in order to do self-supervised training.

Remember that a lot of people have no internal monologue, and even for those that do, it's silent most of the time. They might hope to get that, but the typical case is for there to be no text or voice to extract.

Right but if someone was given a task, internal monologue or not, there will be some brainwave pattern to map thoughts to typing, right? Otherwise, how is the person typing? For a sufficiently complex task, I don't think you can rely on reflexes alone. If I understand the data collection method they're using, they will sit someone in front of a keyboard, give them some task, and collect both what they type out and the associated collected brainwaves.

Sure, but remember the use case being discussed is interrogation. You have a tool that can collect what someone is trying to type, but they aren't trying to type.

And, my point was, there may be zero words to collect at all, because they'll tend not to be actively thinking about any words.

Right, I forgot the initial context in my replies. My bad!

Reply on news.ycombinator.com