mhalttu
Jens OhligI wish more people understood that "I want the computer to generate a natural language text that sounds like a plausible answer to a question about x" and "I want the computer to answer a question about x" are two very different problems.
Odradek@johl könntest du das so kompliziert und ausführlich sagen, daß mehr leute das lesen?
Crumbs the Cat@johl To get people to appreciate this distinction when communicating with computers you would first need to figure out how to get them to appreciate this distinction when communicating with *other people*.
Martin Vermeer FCD
Dr. Angus Andrea Grieve-Smith@johl Now that I think about it, did the TREC conferences ever have a task where there was no valid answer in the knowledge base?
Ok Guy Official@johl the former is sufficient to replace a lot of jobs though. AI is simultaneously over and under-blown. It’s not the singularity or consciousness or anything like that, but that doesn’t mean it isn’t going to be a transformative technology that economically displaces a whole lot of people.
sabik@bulkington @johl
A lot of unkind comments could be made about fields of endeavour where the two are equivalent
A lot of unkind comments could be made about fields of endeavour where the two are equivalent
OkazuYuri@johl I have lost no less than 3 jobs trying to explain this to people.
Jonathan Hartley's old account@EricaFriedman @johl slightly meta, it sounds like you had identified your job as performing task B, while your employers wished for someone to perform task A.
@EricaFriedman @johl A common situation when engineers are in a consulting role.
Brad Martin@johl people who make computers behave in the former way are probably less likely to make that distinction in their own behavior (bullshitter tech bros)
jlapoutre@johl Well I guess that holds for people too, populist politicians?
David Mitchell 
Chat GPT sets off all my teacher alarm bells… it’s the student who is completely baffled but is too proud or frightened to admit it and so desperately tries to figure out the answer they think the teacher wants to hear.
Piper Wrenj@johl It certainly doesn't help that organizations offering the former often advertise it as the latter
Phil Stevens 
@johl I'd also like to see "computer provides primary sources for its answer to question about x" while we're at it.
Even a Wikipedia page would be a nice start.
Mark Rotteveel@phil_stevens @johl You can try to do that right now, and it will simply make up its citations.
Gregory P. Smith (he/him) 
🚲🦝 
@johl I suspect only the minority of people forced to live with a pathological liar in their life are likely to understand this.
Jeff Fecke
Miss Megaphon@johl thank youuuuuu!!!
Alan James@johl yes. i do hope the future versions of this software start including a rating of their confidence in the truth of their answer with every response.
William Gunn@johl Yes, and when we have an AI that can do the second, it can also choose to do the first, i.e., lie.
Nate Gaylinn@johl Totally. It's hard, though! People seem to be hard wired to perceive language as communication.
Raj 🇬🇧🇪🇺💻🖥(🌻🇺🇦;🇵🇸)@johl Orders of magnitude difference.
Wandering Star@johl Yeah if we we could get a robot to spit out an answer that sounds like some hinky 1960s sci-fi robot but it was correct, that would be pretty cool.
Ed Ross@johl
I sometimes go with "I'd like you to generate text about something I know doesn't exist but you'll treat as if it does, just because I told you about it" eg https://developing-baboon-a16.notion.site/Nigel-Farage-s-book-that-doesn-t-exist-yet-ca1b0ce5241747c5a2cdb72cd5994b9f
I sometimes go with "I'd like you to generate text about something I know doesn't exist but you'll treat as if it does, just because I told you about it" eg https://developing-baboon-a16.notion.site/Nigel-Farage-s-book-that-doesn-t-exist-yet-ca1b0ce5241747c5a2cdb72cd5994b9f
Rachel Anne Williams (she/her)@johl philosophers have been debating the nature of knowledge and what constitutes an "answer to X" for literally thousands of years without any consensus or finality. It seems naive to expect computer programmers, let alone the computer itself, to solve the *fundamental* philosophical challenge of epistemology itself when humans themselves cannot do this without enormously complicated social systems like science, which itself is an imperfect and oh-so-human enterprise.
Ben Aveling@johl ChatGPT disagrees.
My colleague, quite accurately, describes ChatGPT as a "glorified paraphrasing engine."
Cornelia Östreich@johl For B, the computer would have to apply scientific methods; for A, it just has to follow the trodden paths.
@johl I'm reminded of a casual comment be a friend years ago, about human psychology, how it feels like our brains are general-purpose thinking machines, but like the rest of our bodies they are actually a combination of disparate systems, each performing a specific function, pressed into service by the exaption of mechanisms that were originally evolved for something else. 1/2
@johl Although comparisons of language models to human psychology are rarely useful, in this case, to my layman's ears, it helps me grasp the idea that "thinking" is not an atomic activity, but can be split into very distinct mechanisms and modes of operation.
Resuna@johl I haven't ever really been a fan of the "Turing Test" as a real thing one should try to achieve. Deceiving humans is super-easy. Even Eliza managed it in some cases.
grob (teeth era) 🇺🇦🏳️🌈🏳️⚧️@johl colleague showed me some vacation planning AI who 1) gave distances between car charging stations which were off by a factor 3 -> battery dead 2) recommended non-binary couples to visit UAE 3) gave women's travel advice along the lines 'just go, it'll be fine'. 😵
Adam Isacson@johl what’s the first one useful for, besides “entertainment?”
David J. Atkinson@johl Now, substitute the word “student” for “computer.” It still makes sense!
Derek Bruff@johl Yes! Very well put.
Droid Boy 
@johl The difference between a language and a knowledge model.
Greg Johnson@johl many computer users have hoped for an answer to wanting the computer to generate a natural …
MWibralSo true, indeed.
Iris Volk@johl The computer doesn't understand, and people don't understand, that the computer doesn't understand. We're stuck. 🤷♀️
Mostly Harmless (Cam Johnson)@johl and that “sentient being” vs. “string of algorithms that mimic what a sentient being might say” are different too.
MatthewPCooke@johl In a large number of contexts it already gives answers more likely to be correct than a on average human. And it’s far faster and more reliably than using search in many contexts. One has to factor in how good the alternatives are when judging.
apm77@johl I recently watched a Youtube video where the top rated comment was someone sharing "research" they'd done via ChatGPT, and the channel owner endorsed it as interesting. *No-one* pushed back and advised caution. It was a science channel, so they definitely should have known better.
Hanno Terbuyken@johl Many times this.
teledyn 𓂀