Similar, organized by Polish students, the P.I.W.O. project:
> Transforming their dormitory building into a light show extravaganza, the students at Poland’s Wroclaw University of Technology demonstrated their tech-savvy skills with this large-scale installation of pixel-like flashes set to an equally animated soundtrack. Called “Projekt P.I.W.O.,” [Potężny Indeksowany Wyświetlacz Oknowy] (the acronym means “beer” in Polish), it’s simultaneously humorous and beautiful—particularly the Michael Jackson tribute about seven minutes in.
NotebookLM is great to get an overview of a publication.
I created a short podcast focusing on HCI publications using NotebookLM
https://www.deep-hci.org/
Just posted some ISWC, MobileHCI and UbiComp papers, UIST is up next.
"GPT-4 and professional benchmarks: the wrong answer to the wrong question
OpenAI may have tested on the training data. Besides, human benchmarks are meaningless for bots."
ha ha ... every human is a fool in that case (as every human holds some unreasonable believes and opinions some point in time).
"He has discredited his mind?" This sentence does not make sense to me. Explain what you mean by this statement.
I found his NYT piece well argued. How can you fix any language model similar to ChatGPT to prevent the mistakes shown in the article? (I don't think you can).
He makes irrelevant and baseless claims. First we don't really know how the mind works and therefore we really have no idea if the language models are path toward AGI or not. He really doesn't have an idea how a child obtains a language. You really need a lot of data to acquire grammar, no minuscule data is not enough. I have observed it personally with my multilingual children, their mother tongue, to what they have the least exposure and what is the most complex, is not still not fluent for them. His claims are simply false, easily refuted by empirical data.
ok, you didn't answer my first questions. I try a couple of more.
your first argument is useless, as anything might be a path to AGI ... first order logic / expert systems are ...
regarding empirical data, you are bringing up anecdotes that don't hold up. Humans can often learn a concept from 1-2 instances (take the face of Albert Einstein) (even your kids can do that in grammar :) , maybe check how language models are trained in comparison.
do you have any references for your claims? How much data does a human need to learn grammar? do you have any estimate?
you say you can refute Chomsky's claims by empirical data. please show me that data/studies, don't talk about anecdotes from your kids.
"His claims are simply false, easily refuted by empirical data."
he underlines his claims with examples showing what he means. I recommend you to read the original article by Chomsky, as you seem to have a problem understanding his arguments.
Very skeptical about this. I'm working with heart rate monitors and different physiological signals. The wrist-based smart watches are still sometimes way off compared to an ECG baseline, depending on skin type and how they are worn.
Even step counting with them is not accurate at all.
Can't imagine that Blood Glucose tracking will work anytime soon for a large population.
the models we have today will never be able to read .. they are just able to produce something that we cannot distinguish from human output.
we should be more careful on how we use them. in my opinion, ChatGPT and similar will be horrible for search on the web, as we are flooded with text that looks like a human wrote it but it does not add any knowledge or new insight.
That is because it has seen literature or code that is syntactically similar enough to what you give it, so that it can autocomplete its way to a semi-intelligent response.
Then we have a different definition of understanding. A good half of these debates come down to definitions at this point.
To me, ChatGPT clearly displays degrees of understanding. Often when I ask it to do things like make a certain modification to a piece of code, I'll ask it "why did you choose to do it like that?" and it gives a coherent explanation. That it correctly did what I asked and that it's explanation of why it did that way comport with one another implies understanding in my view. How could it not? What else would that imply? And if I ask a human to do the same thing and get the same result you'd have no problem with saying "well it's able to do that because it understands, and if it didn't understand then it would have zero capability of doing that". So, what's with the double standard?
For me, it's matrix multiplications on data made to output things that are approximations to what a human has written before on that question or text you're prompting it.
I know how training a deep neural network works and there's no reading involved for me :)
It's just estimating what a person would write (being trained on a really large data set of what people wrote). If it's trained on gibberish, it will just output gibberish. If I ask you to read a book with gibberish words would you do it? ChatGPT would (according to you) "read it" and recite the gibberish I gave it without a problem.
The matrix multiplications used to build these models don't show any comprehension, agency, or consciousness. Their point is to estimate the next data point given the previous data point, they will perform the same task independent from which training data we give them. That's not reading.
I don't say this might not change in the future, yet treating ChatGPT like a human using words like "reading," "understanding" etc. or intelligence for that matter is just anthropomorphizing for me.
I look at it from a general system theory point of view first and foremost as that's a level of abstraction higher than both humans and machines, yet you can find clear sets of definitions that describe all of these things. They go so far as to include knowledge, understanding and degrees thereof. I've started working my way through "On Purposeful Systems" as I've been finding most people's definitions far too hand wavy to be useful in thinking about the topic. That's been quite interesting.
One thing that stands in my mind from having read that is that the structural properties of the system matter in terms of its possible function, and I'm referring to the exact structure that includes the model trained on the non-gibberish text and treating that as the specific system I'm referring to. It seems when you say ChatGPT you're referring to the entire functional class of objects that implement the transformer architecture, which I'm not.
After thinking about it for a while the second thing I would note is that you're focused only on it's intrinsic function of predicting the next token, which as you say it always performs that function regardless of the training data, and I agree with that. Where we differ, I think, is I'm considering the specific structure which includes the model trained on non-gibberish data as an instrument (not a human!) that we use for it's extrinsic functions. Normally an individual or system's extrinsic functions are so narrow they're a lot easier to define, and we've never interacted with a system until now that the extrinsic function is like "understand what I say and reply accordingly". Think about it from the perspective of would you use or have any utility for GibberishGPT? No, same way I wouldn't read a book of pure gibberish.
to make my point easier: for my definition of to read, to understand and comprehend you need a body, an embodied mind. you need agency and consciousness. ChatGPT has none of those, it's an implementation of several matrix functions.
the specific model you are referring to does nothing, it's not acting (it's not reading anything). you have to press enter for it to work. so it's not reading, you (and OpenAi engineers) are doing its actions for it.
I'm wondering about the study ... I'm a complete novice so might get things wrong, yet the participant description has not much information (e.g. average, std of the age, gender etc.). N is also just given in the abstract.
Also the study does not mention how the experimental setup was:
"Subjects were administered a comprehensive psychometric battery of fluid and crystallized intelligence tasks,"
There can be substantial differences, just on how the experimental setup was administered (e.g. potential ordering effects between the tests) and when the scans were done.
Does anybody know if that's a standard procedure to determine "general intelligence"? Sounds vague to me.
Reading science papers (my field is biochem, YMMV for other disciplines) is a lot of work if you're not already expert in its very specific field and topic.
Paper authors try to be as succinct/brief as possible, and in my experience editors generally encourage this as well. There are incentives to do so.
In order to understand what that part of the paper means, you would need to actually read every reference, and maybe even their references, and maybe even go up pretty far up the reference tree. You might try and take a shortcut by reading recent or well-cited review papers on that specific subject. In order to take a shortcut from that, you might try and consult a handful of textbooks.
If you were an expert in the field, you probably wouldn't need to refer to the textbooks, and you might even know the rough ideas behind enough of the papers and might check only a few of the references to be sure.
Then beyond this, any variation from the afore-cited methods will usually be detailed within the paper, possibly in further sections or in figures or tables (and from what I can see, that seems to be generally the case in this paper).
Being succinct is also a neat excuse for leaving out details that are actually necessary for any scientist to replicate your paper. Oh and it allows certain interdisciplinary works to incorrectly abstract away technical details from the secondary domain.
You're not wrong, and in an ideal world it is a sensible way for experts to communicate amongst themselves. But at the same time this sort of deep reading is a fool's errand at least half the time IME, because the authors of the paper themselves did not carefully read every reference. And even when they did there are often subtle unpublished differences between a cited protocol and the protocol they actually used.
To be fair, it's possible this is a domain specific phenomenon, and I'm not especially familiar with human intelligence research. But it's a real problem in the areas of neurobiology and computational neuroscience I've studied.
I think it's in large part a natural consequence of trying to scale science in the wrong way, but there are also bad cultural factors and a few legitimately bad actors that push the issue even further.
I meant it seriously though. Unless you're a deep expert at something, most likely you're just going to use Java to hook into that thing's SDK or whatever.
OP didn't come across as someone who was trying to learn PhD-level knowledge about a topic. In which case, being good at Java is probably as good a skill as any.
don't know how long she worked on that script (conveying the idea first, details later). it's the first time I really understood the basics of linear dynamics. super impressive.
