Hi, Michel Verdier wrote: > I think they need better intelligence to better use their data.
Actually i think that current AI lacks of disciplined and good willing reasoning rather than of IQ. It is astounding how far ye olde Perceptron made it meanwhile by being augmented with oddities like non-linear voodoo layers or 8-bit floating point number arithmetic. Remembering the intelligence tests of my younger years i'd expect to be beaten by AI on many of their topics. But as small the space of combinations of xorrisofs arguments is compared to the size of an AI parameter space, the AIs are still not able to correcty answer a question like "How to modify a bootable ISO of distro XYZ ?". (I pick up the debris of their flawed answers in the internet.) Joe wrote: > The problem is that 'AI' is a hoax, there is no 'I'. What we have now is > ELIZA with about a trillion times as many computer resources, but not a > single bit more actual intelligence, since we don't know how to make > that. We only know one way to make human intelligence and it is quite similarly obscure as AI training if we consider the ~ 3.5 billion years of evolution which enabled our mass production of humans. And many of them will never qualify for what we as computer oriented people are undisputedly willing to call "intelligence". > It's a large-scale expert system that hasn't been trained by experts. Again similar as with humans: Those who can, do. Those who cannot, teach. Those who cannot teach, teach teachers. I recognize in AI lots of deficiencies which i first saw in journalism and academic communities. Form beats meaning. Word beats structure. So what is missing in my view is commitment to "Nullius in Verba". AI which does not believe in everything that its masters give it to read. Of course we have to be aware of D.F.Jones' novel "Colossus" which stems from the same time as the Perceptron. Andy Smith wrote: > If an LLM were asking for help, and I knew it was an LLM, then I would > also know that any time I spent on responding would be me donating my > free labour to a huge corporation for it to make money off of It is my considered decision to allow exploitation of my personal sport. I'm not really in the situation of https://xkcd.com/2347 but i've seen my hobby horse in use by organizations which surely would not have hired me 20 years ago. I don't begrudge them their profit. Making money is hard work in itself and it leaves own scars on the soul. > in what is likely to be a highly damaging bubble. [..] > It is perhaps a little illogical since it doesn't seem to bother me what > the human would use the knowledge for, while it does bopther me what the > LLM uses it for. This is an interesting aspect of free software work in general. What if somebody does something really evil by help of my stuff ? Am i responsible ? Should i scrutinize people for possible bad intentions before giving them advise ? (The GPL of inherited code forbids me to impose the demand for being not evil, even if i could give a convincing definition of what i mean. But that is a rather lame excuse, ethically. I could get rid of that code and then start a license crusade.) > Soon we'll have our own agents offer > to take away the tedium of doing that by doing it for us. I'm on the side of the non-evil ones. :)) [email protected] wrote: > I think the point is that the currently dominant "AI" shops > aren't about facts. There's not much money in that. But there will be when the old experts retire and the rich people need a doctor who knows the job. > There is "money" (actually potential, speculative money) A very important point. For now AI is predominantly expensive and gluttonous. I wrote: > > [...] so we get smoothly into the > > pampered and isolated state of the Spacers in Asimov's novel > > "The Naked Sun". [email protected] wrote: > No, no. As much as I admire Asimov, it's more Harry G. Frankfurt [1] > here. Sounding truthful is the aim for them, truth is just > uninteresting. But it would be tremendously useful and monetarily valuable to have a simulation of a good willing rational expert. For now AI simulates highly educated imposters. I deem it surprising that the art of imposting was so easy to acquire. Now, if the swindler would discover its love for honest science ... Have a nice day :) Thomas
