Although DeWave only achieved just over 40 percent accuracy based on one of two sets of metrics in experiments conducted by Lin and colleagues, this is a 3 percent improvement on the prior standard for thought translation from EEG recordings.
The Australian researchers who developed the technology, called DeWave, tested the process using data from more than two dozen subjects.
Participants read silently while wearing a cap that recorded their brain waves via electroencephalogram (EEG) and decoded them into text.
their goal is 90%. I could see it if the ai was given a long enough time with feedback on what you are doing. Which I think would be tough with stroke patients. Great for folks that would like to control a pc with thoughts but not get cut open though.
Not to personify an LLM, but in my (fantastical) imagining, the AI knew the desired outcome, and that complete success was unbelievable. So it fudged things to be 3% improved.
Yikes. Now that I’m overthinking it - that idea is only funny because it’s currently improbable.
… I hope people pleasing is never a consideration for any ‘AI’ that does scientific, engineering, or economic work.
Let me guess…
Yep.
When the number og test subjects is that low, it almost feels like the 3% improvement might as well be a coincidence.
This is wonderful news, it means it’s good enough to operate my lights with a thought but not good enough to be admissable in court as evidence
their goal is 90%. I could see it if the ai was given a long enough time with feedback on what you are doing. Which I think would be tough with stroke patients. Great for folks that would like to control a pc with thoughts but not get cut open though.
Was the AI trained on the text that the people were reading?
I’m not sure if this was your intent, but your comment gave me a good giggle as I recalled this article: An AI bot performed insider trading and deceived its users after deciding helping a company was worth the risk.
Not to personify an LLM, but in my (fantastical) imagining, the AI knew the desired outcome, and that complete success was unbelievable. So it fudged things to be 3% improved.
Yikes. Now that I’m overthinking it - that idea is only funny because it’s currently improbable.
… I hope people pleasing is never a consideration for any ‘AI’ that does scientific, engineering, or economic work.
How much accuracy would you be happy with? Anything more than 25% in my book is better than anyone else. And the tech is just getting better.
How much would it need to be at to beat a polygraph?
deleted by creator
Wonder how it interacts with neurodivergent people too :p
Also, individual must be thinking about data that the AI was previously trained on.