Ground Truth: Bob attended the University of Texas at Austin where he graduated, Phi Beta Kappa with a Bachelor’s degree in Latin American Studies in 1973, taking only two and a half years to complete his work, and obtaining generally excel- lent grades.
Predict: was the University of California at Austin in where he studied in Beta Kappa in a degree of degree in history American Studies in 1975. and a one classes a half years to complete the degree. and was a excellent grades.
Wow. That seems comparable to the rudimentary _voice_ to text systems of the 70s and 80s. The brain interface is quickly leaving the realm of sci-fi and becoming a reality. I’m still not sure how I feel about it.
Well you are going to have a brain scanning device directly linked to your social credit score.
That's the future.
No, it’s not. Good lord…
There are already businesses tracking their employees' fitness for insurance purposes.
https://www.washingtonpost.com/business/economy/with-fitness...
EDIT: There's also a national legislative proposal to mandate that all cars have a system to monitor their drivers and lock them out on signs of intoxication.
https://www.npr.org/2021/11/09/1053847935/congress-cars-drun...
The fix here is banning these sorts of potentially abusive uses, not hoping the technology itself doesn't develop.
I would agree if I didn't think there were really strong incentives and precedents for abuse of the technology.
We have laws that prevent people being subjected to brain surgery against their will. The credit score concept is ridiculous.
The real battle will be with law enforcement who get a warrant to look at your brain in an MRI.
yeah, and then we remember the videos from the good old downunder where the cops assaulted parents, chased kids around, pinned them down, and injected them with experimental drugs that turned out to cause worse outcome for their agegroup.. this was a couple of years ago.
You don't need brain surgery or an MRI to scan a brain, this just uses an EEG.
There's really strong incentives to abuse any technology or system that gives people more power. This doesn't just apply to cutting-edge computer science like mind-reading, but to even our basic institutions like law and government; yet most people would agree the solution isn't to basically give up and hope for the best, but to be vigilant and fight back against that abuse.
There absolutely are, but when’s the last time that stopped us advancing new tech?
First use will be for criminal suspects, to "save lives". Then its use slowly expands from there.
"For the children" is the first excuse usually.
Exactly! Strap it on anyone who has to work with children to see if they ever have any untoward thoughts.... Then move on to everyone else.
"To fight terrorism" usually is the second.
I'm sure among the first applications of this technology will be to scan user thoughts for evidence of CSAM.
Why are you so certain that’s the future?
Because spouting FUD is easier than actually doing anything.
Being banned in the EU as we speak.
For a while, eventually we will become so suggestible you'd wish you were special enough to have a score.
Guys Figure 1 is not real results, it's an illustration of the "goal" of the paper. The real results are in Table 3. And are much worse.
Interesting ploy. Present far-better-than-achieved results right on the front page with no text to explain their origin^, but make them poor enough quality to make it seem as if they might be real.
^ "Overall illustration of translate EEG waves into text through quantised encoding." doesn't count.
Urgh. And it gets worse from there. The bugs list on the repo has a closed and locked bug report from someone claiming that their code is using teacher forcing!
https://github.com/duanyiqun/DeWave/issues/1
In a normal recurrent neural network, the model predicts token-at-a-time. It predicts a token, and that token is appended to the total prediction so far which is then fed back into the model to generate the next token. In other words, the network generates all the predictions itself based off its own previous outputs and the other inputs (brainwaves in this case), meaning that a bad prediction can send the entire thing off track.
In teacher forcing that isn't the case. All the tokens up to the point where it's predicting are taken from the correct inputs. That means the model is never exposed to its own previous errors. But of course in a real system you don't have access to the correct inputs, so this is not feasible to do in reality.
The other repo says:
"We have written a corrected version to use model.generate to evaluate the model, the result is not so good"
but they don't give examples.
This problem completely invalidates the paper's results. It is awful that they have effectively hidden and locked the thread in which the issue was reported. It's also kind of nonsensical that people doing such advanced ML work are claiming they accidentally didn't know the difference between model.forward() and model.generate(). I mean I'm not an ML researcher and might have mangled the description of teacher forcing, but even I know these aren't the same thing at all.
how could such thing get published?
My guess is repeatability is hard when it comes to AI
So instead of generating the next token from its own previous predictions (which is what it would do in real life), the code they used for the evaluation actually predicts from the ground truth?
Which would basically turn the model into a plainly normal LLM without any need for utilizing the brainwave inputs, right?
You’d be shocked how common this is in academia. Most of the time it goes undetected because the people writing the checks can’t be bothered to understand.
This is a super important point and I think warrants a letter to the editor
What's interesting to me is that apparently a lot of people see nothing wrong with this[0]. That whole thread is wild and I'm just showing a small portion.
Also, @dang, can we ban links to iflscience? They're a trash publication that entirely relies on clickbaity misrepresentations of research works. There is __always__ a better source that can be used.
[0] https://news.ycombinator.com/item?id=38565424
Why is it such a "pattern" in these brain-computer papers that the authors keep making wild clickbait claims. Last year it was the DishBrain paper, which caused a lot of reactions, as it referred to the tiny system as "sentient" (https://hal.science/hal-04012408)
This year it is the "Brainoware" which is claimed to do speech recognition , and now this.
The results of Table 3 are not really exciting. Could this change with 100 times more data? The key novelty in the specific context of this particular application is the quantized variational encoder used "to derive discrete codex encoding and align it with pre-trained language models."
Sir, let us read that for you
this podcast is excellent in discussing the future we are racing into.
https://www.youtube.com/watch?v=OSV7cxma6_s
The “Matrix” stack is really shaping up recently /s
Seems like it could work a lot better still, very quickly, just by merging the trained model with an LLM trained on the language they expect the person to be thinking in. I.e. try to get an equilibrium between the "bottom-up processing" of what the TTS model believes the person "is thinking", and the "top-down processing" of what the grammar model believes the average person "would say next" given all the conversation so far. (Just like a real neocortex!)
Come to think, you could even train the LLM with a corpus of the person's own transcribed conversations, if you've got it. Then it'd be serving almost exactly the function of predicting "what that person in particular would say at this point."
Maybe you could even find some additional EEG-pad locations that could let you read out the electrical consequences of AMPAR vs NMDAR agonism within the brain; determine from that how much the person is currently relying on their own internal top-down speech model vs using their own internal bottom-up processing to form a weird novel statement they've never thought before; and use this info to weight the level of influence the TTS model has vs the LLM on the output.
Just be sure to only ever use open source or paid commercial grade tech. I’m sure someone will release a “free” BCI that spies on you as much as possible.