Mastodawn

saxnot ➡️ GPN, RockHarz Apr 23

@kohnzn LLM are pretty bad at facts.
but gettings inspiration for text which I then not copy and paste is pretty great.

LLM have been invented to master languages (specificially: translating them) and I like that about LLM.

But I am still very reluctant of sending data to companies I don't trust

@saxnot @kohnzn Apart from the syntax processing parts, which are what fool people into thinking the things are ‘intelligent’, they are just very advanced naïve pattern processors akin to Spam Assassin.

ChatGPT, for instance, simply finds things in its databases that are near each other and strings them together. I know this because it once answered a question with my answer for it but giving my opponents’ retort as the justification! So these were near each other in the databases.

@saxnot @kohnzn

But it is easily proven that LLMs are a dead end for ‘artificial intelligence’.

What is a human being? An organism that is almost the same as a chimpanzee, with an overall intelligence that is very, very similar to a chimpanzee’s. But does a chimpanzee have language? A chimpanzee can communicate, but LANGUAGE is infinitely recursive in structure. A CHIMPANZEE HAS NO LANGUAGE.

Thus human intelligence CANNOT be modeled on manipulation of language. It just CANNOT.

@saxnot @kohnzn

This is why LLMs are stupid.

They do not ‘hallucinate’. They do none of the things people commonly say they do. They do exactly what you would expect if you view them as naïve pattern processors with syntax processors as front-ends.

@saxnot

Here BTW is the problem (not original to me) that I posed to that LLMs. They cannot answer correctly, except that once when ChatGPT happened upon my answer. Humans almost always also get it wrong, but I am sure my answer is the correct one. I have thought out the philosophical details.

‘With what probability is six the final digit of pi.’

ChatGPT gave my answer zero and justified this by claiming the problem was ill posed. That is the DETRACTORS’ claim, saying there is no answer!

Barry Schwartz 🫖

@saxnot (But, if the detractors were correct, then the question ‘Does pi have a final digit?’ ALSO would be ill posed and have no answer.

In fact, ‘Does pi have a final digit?’ is equivalent to ‘Is the probability x is the final digit of pi equal to zero for x = 0,1,...,9?’, so, again, my original problem is NOT ill posed.

Even HUMANS cannot solve this problem if they do not go beyond mere shallow verbal reasoning and instead VISUALIZE the problem. I have PICTURES in my head.)

saxnot ➡️ GPN, RockHarz May 1

@chemoelectric I agree a naive answer might be similar to 1/9 but I have the impression the LLM answer is correct because pi does not have a last digit.

In our decimal system the ratio pi is just a wonky number. Of course if you have a system more inclined to circles and such you might have a less irrational number

in decimal represention pi goes on and on. There is no last number, thus asking for it makes no sense. It's like asking "what was before the big bang?" when time itself started then

@saxnot We are not asking for the last digit of pi, we are asking for the probability that 6 is the last digit of pi. There is no last digit of pi, therefore the probability is zero.

That is the reasoning behind the answer. The reasoning is that there is a theorem proving there is no last digit of pi. The reasoning is NOT that the question was ill posed!

@saxnot I have an anecdotal relationship with pi in that in the mid-1980s I met the Chudnovsky brothers who are famous for finding digits of pi. That is pretty much the extent of the anecdote. We didn’t discuss pi.

As for the Big Bang, I am agnostic on that. I do not believe ‘big’ stuff physicists tell us, just because they have PhDs in Physics. I know for sure that Clauser, Aspect, and Zeilinger were awarded the 2022 physics Nobel prize in error. Their work is garbage. (See my pinned toots.)

@saxnot Go look up ‘Bertlmann’s Socks and the Nature of Reality’, John Bell’s paper is the classic in the field.

(Clauser has become an infamous climate denier BTW. No shocker if he also does not understand thermodynamics. He is an awful physicist.)

One gets to a part where he ‘encodes local reality’. At this stage he conflates a function representing causal influence with a conditional probability. This is a clear no-no in random process analysis.

So the entire field is garbage.

@saxnot There is really no such thing as a fundamental physicist today. They are pseudoscientists who cannot correctly do mathematics and do not understand what a ‘theorem’ is.

Bell’s paper is freely available in its manuscript form, with a few minor errors.

Much of the paper is handwaving babble.

There are some examples of his logic that he gives for which counterexamples are easily contrived. It is dirt simple to contrive a counterexample to the ‘heart attacks’ logic, for instance.