Re: unboxed structure fields (or class slots)
"Tim Bradshaw (as tfb at tfeb dot org)" <[email protected]>
| Newsgroups | gmane.lisp.lispworks.general |
|---|---|
| Message-ID | <[email protected]> |
On 13 Jan 2026, at 19:01, Yuri Davidovsky (as work at disclosure dot ie) <[email protected]> wrote: > > We only know what we have observed in the reality around us Where, in the 'reality around us' was the proof of Fermat's last theorem before 1993 / 1995? Seriously, this is arguing about angels and pinheads. In particular the following three statements apply: 1. If physics is computable, then a computer can, in principle, simulate any physical system. 'Computable' has a fiddly definition but it comes down, I think, to 'you can compute the terms of a Cauchy series convergent to the solution of the appropriate equations'. You also need to assume that there are laws of physics and there are only a finite number of them I think. 2. Neural networks can simulate a universal Turing machine, with the same caveats that you need an unbounded amount of storage that apply to any such simulation. 3. A universal Turing machine can simulate any neural network (with a similar caveat about actually computing terms of a Cauchy series convergent to the actual value). (2) and (3) say there's no difference in power between the two approaches (one may obviously be much better than the other in practice!). (1) says that, if you believe that humans are physically-based (many people do not) then the interesting question is whether physics is computable. If it's not then it's possible that neither approach will work, if it is, then both will. But (2) and (3) say that people arguing that 'neural networks can never x, but some other computatonal system can x' are simply wrong, for any x. So please stop arguing that. --tim _______________________________________________ Lisp Hug - the mailing list for LispWorks users [email protected] http://www.lispworks.com/support/lisp-hug.html