Am Fr, 17. Mär 2023, um 14:11, schrieb John Clark: > On Fri, Mar 17, 2023 at 8:26 AM Telmo Menezes <te...@telmomenezes.net> wrote: > __ > >> *> Well, this is Machine Learning 101. If you train a model, it will always >> perform better* > > Well yes, if a machine couldn't learn then it wouldn't be intelligent!
Ok John, I have an algorithm here that, can 100x times your money every month in the derivatives market. I can demonstrate this by running it on the training data. I will sell it to you for 100K. An incredible bargain, considering that you can become the richest man in the planet in just a few months. Interested? >>> >> When human beings take an IQ test that is almost certainly NOT the first >>> >> test they've ever had, and like GPT-4 humans are also trained on a huge >>> >> amount of data, without it neither you nor GPT-4 would even know how to >>> >> read the questions. >> >> *> Yes, but GPT-4 and human brains are very different things.* > > They are similar in some ways, they both process information, and they are > different in other ways, one processes information using carbon chemistry and > the other processes information using silicon electronics. Other ways in which they are different are also that: one is a Turing-complete self-organizing general learning algorithm with a highly evolved utility function, and the other is a model trained with gradient descent and a huge amount of data to predict the likelihood of words given a context. Huge progresses is being made, but we are not at the human level of generality of intelligence and autonomy. Not even close. I am sure we can get there, but I would argue that too much hype too early hurts the cause. That's how we got the first AI winter. > >> *> GPT-4 has superhuman memory capabilities* > > Yes > >> *> and almost certainly subhuman reasoning capabilities.* > > If so I've seen no evidence of it, I have however seen evidence that the > opposite is true. I fear that you are falling for the very human bias (I fall for it so many times myself) of seeing what you want to see. >> *> I bet that it would take many human lifetimes to actually read the entire >> training datatset of GPT-4.* > > I'm sure that's true. Am I supposed to think less of GPT-4 because of that? No, but you are supposed to remain a scientist and keep applying the same fundamental machine learning principle that brought us all the way to GPT-4: A machine learning system can only be objectively evaluated by applying it to data that was not used to train it. This is the only way to distinguish between true generalization (learning) and over-fitting. >> *Again, it is important to understand what exactly GPT-4 is doing. It is >> certainly impressive, but it is not the same thing as a human being taking >> an IQ test,* > > It's the same thing if you treat both humans and machines as black boxes and > concentrate on what they do. Like it or not that's the only way we can deal > with our fellow human beings that we encounter in everyday life, we have no > way of knowing what's going on inside their head, all we can do is observe > their behavior. Maybe Einstein was an idiot but he just had an ability to > push a pen in such a way that he produced brilliant physics papers, but > nobody believes that; instead we would say if somebody could write physics > papers that were as brilliant as Einstein's then that person would be as > smart as Einstein. > > It's interesting that until a few years ago the Turing Test was not very > controversial because most thought it would be centuries before a machine > could pass it, and many proclaimed a machine would never be able to pass it, > but now that a machine has indeed passed it they say the Turing Test is not > important, even though they personally still use the Turing Test a 1000 times > a day whenever they judge the conscious state of one of their fellow human > beings. Actually if GPT-4 really wanted to fool somebody into thinking it was > a human being it would have to dumb itself down. > I do think that passing the Turing test is impressive, although it is true that most AI researchers never took it very seriously, it was more of a pop-science thing and a niche interest of the chatbot community. Anyway, I agree with you that a major milestone has been achieved. GPT-4 and image generators are a type of intelligence that we had never seen before. Maybe the first time such a thing arises in this galaxy or even universe, who knows... They are probably also similar to stuff that happens in our brain. But what they are not is something you can be compare to a human mind with an IQ test in any meaningful way. That is just junk science. Telmo > > > John K Clark See what's on my new list at Extropolis > <https://groups.google.com/g/extropolis> > > h5z > 8bm >> -- You received this message because you are subscribed to the Google Groups "Everything List" group. To unsubscribe from this group and stop receiving emails from it, send an email to everything-list+unsubscr...@googlegroups.com. To view this discussion on the web visit https://groups.google.com/d/msgid/everything-list/1c34a5ec-a2f2-4d6e-a3c7-fb44d32ebe45%40app.fastmail.com.
