OR, I could translate the entire previous email into Norwegian, just to see
how well it translates. I told it to translate it to "conversational
Norwegian" and that it should mimic as much of the English poem's form as
possible. Of course, I can't read Norwegian, so I have no idea how well it
did. So... Pat. How did it do? It looks like it only sorta-rhymes, and
not very well at that.
---
Han har helt rett. ChatGPT, eller nesten enhver stor språkmodell, vet
ingenting. Det kaster bare pil på et veldig komplisert dartbrett. Du kan
ikke stole på noe det sier. Men det betyr ikke at det ikke er nyttig. Det
er VELDIG nyttig.
ELLER skal jeg formulere det slik:
Han treffer blinken, i det han sier,
En sannhet, alle kan forstå med en gang,
At store maskinmodeller, med mye å spare,
Er som pilkast, på et stort brett de fanges.
Dets tanker, selv om nyttige, kan ikke være tillitsvekkende,
Det er bare gjetninger, ikke noe som er planlagt,
En tråd i et stort nett, de ikke kan bære,
En farlig lek, å stole på det de har skannet.
Verdien skal likevel ikke bli avvist,
For i denne verden, er det en hjelpsom hjelp,
Med å hjelpe oss med svar uten splitt.
Og en verdifull ressurs, aldri overbetalt.
Og selv om kunnskapen ikke alltid er alliert,
Er verdien fortsatt, et verktøy vi har tillit til.
Og jeg måtte faktisk kjempe litt med ChatGPT for å produsere dette. Noen
ganger går ChatGPT fint, og plutselig mangler en vers en linje, og de andre
linjene rimer ikke, og jeg vil til og med be ChatGPT om å verifisere at det
oppfyller kravene mine, og det vil si "ja" og fortelle meg nøyaktig hvordan
det oppfyller kravene selv om det ikke gjør det. "Vers 1 rimer ved hjelp av
ABAB riming skjema". "Nei, det gjør det ikke". "Unnskyld feilen min, her er
et nytt vers 1 som gjør det ..."
Du må jobbe med det, gjøre litt frem og tilbake redigering, etablere et
forhold. Det er ikke en maskin.
Scott
---
On Sun, Feb 12, 2023 at 8:39 AM Mr. Ransom <mr.ransom@xxxxxxxxx> wrote:
He's totally right. ChatGPT, or pretty much any large language model,
doesn't know anything. It's just throwing darts at a very complicated
dartboard. You can't trust anything it says. But that doesn't mean it
isn't useful. It's SUPER useful.
OR should I word the above as...
He's on the mark, in what he does declare,
A truth, that all can plainly understand,
That large machine models, with much to spare,
Are but like darts, at a vast board they’re fanned.
It's thoughts, though useful, in true trust can’t ensnare,
It's learnings but guesses, not at all planned,
A strand, in a vast web, it cannot bear,
A dangerous game, to trust what it has scanned.
It's worth, however, should not be derided,
For in this world, it’s a helpful aid,
Assisting us with answers undivided.
And a valuable resource, never overpaid.
And though its knowledge may not be allied,
It's worth remains, a tool on which we confide.
And I actually had to fight ChatGPT a little to produce that. Sometimes
ChatGPT will just roll along doing great, and then suddenly one verse is
missing a line, and these other lines don't rhyme, and I'll even ask
ChatGPT to verify that what it's done adheres to my requirements and it'll
say "yes" and tell me exactly how it meets the requirements even though it
doesn't. "Verse 1 rhymes using the ABAB rhyming scheme". "No it
doesn't". "Pardon my mistake, here's a new verse 1 that does ..."
You gotta work with it, do some back and forth editing, establish a
rapport. It's not a *machine*.
Scott
On Sat, Feb 11, 2023 at 10:49 PM Steve Crane <steve.crane@xxxxxxxxx>
wrote:
https://boingboing.net/2023/02/10/steve-wozniak-says-chatgpt-is-useful-but-it-can-make-horrible-mistakes-video.html
Steve Crane
about.me/stevecrane
