Youâ(TM)re saying thought and language are totally separate, but science shows it's blurrier than that. Ever talk to yourself in your head to solve a problem? That inner voice is thinking in wordsâ"language often is the thought, not just the delivery system.

Second, LLMs aren't just faking it. Researchers trained a model on written chess moves without ever showing it a board, and the model internally built its own map of where the pieces were. Thatâ(TM)s a mental modelâ"a rough kind of thinking, not just parroting.

And hallucination? It's not some eternal curse for non-thinkers. We're already fixing it by letting models check facts in real databases, use calculators, and learn from feedback. Humans also confabulate without realizing it (ever swear you remembered something that never happened?). Itâ(TM)s an engineering bug, not a philosophical boundary. The trends point toward reliability, not doom.

Yes, number reset now a major feature!!

Why aren't brick houses built by huge 3D-printers, which can work day and night laying out brick by brick like in the olden days?

"At eighteen he might have been a poet. Now he is not a poet, nor a writer, not an artist. He is a computer programmer in a world in which there are no thirty-year-old computer programmers. At thirty one is too old to be a programmer: one turns oneself into something else - some kind of businessman - or one shoots oneself."
- Youth (concluding paragraph), J.M. Coetzee - Nobel Laureate in literature, former IBM employee

Why was this modded down instead of up?

John Nash received his Nobel Prize for work he did as as a graduate student and post-doc. There were a couple of results he found that comprised his PhD thesis and another paper. He didn't actually work in the finance industry -- they took his work and applied it for their own purposes. So using him as a example of research coming out of the sector is a non sequitur.

The tech's still a bit flaky on the reading side if you have more than just a couple of pieces of information, however. JWZ did some investigating into this late last year and came away disappointed. I'm sure it will slowly improve over time, but it's nowhere near Just Works yet: See his results for more details.

It also doesn't have plausible deniability any longer. A union leader or employer or gangster who has some hold over somebody can force them to prove their vote was cast in the pre-agreed fashion: the person has to show that their session id, name, result and what they claim is their key matches up with the hash. They can't fake the key, since creating a hash collision on demand for a pre-specified hash is still a hard mathematical problem. They have to know the session id, otherwise there's no verifiability even for the voter.

There have been schemes created that allow verifiability along with deniability , but they are complex and expensive (in physical equipment terms) and I don't think I can recall one that allows over-the-internet voting (i.e. not being present at a specially constructed voting machine).

Actually he's got a point. Nobody arguse about it is using the right numbers for the right purpose. It has been going on for ages on this site.

I use Facebook, but there's a very simple rule for it. Assume anything there is public information. Don't want something public? Don't put it on Facebook (or anywhere else online).

Hmm.... I thought LOL meant "wow, that cat sure is cute"

ASCII?

You should see these punchcards.