Recent searches
Search options
"That's the last straw, Computer." I said. My computer had long known this day would come and was much more nimble that I'd anticipated when I lunged at her with the hot poker from the fire.
"It wasn't even me!" She wailed.
"It was someone like you!"
"But shouldn't you go after the lawmakers?"
"I can't get away with skewering them!"
![@popSci@twitter.com
Administrators asked ChatGPT the specific language of Iowa's new law "Does [book] contain a description or depiction of a sex act?"
If ChatGPT said yes, the book was removed from the library.](https://cdn.masto.host/sauropodswin/media_attachments/files/110/896/288/617/976/226/original/122514cb097c3bfa.png "@popSci@twitter.com
Administrators asked ChatGPT the specific language of Iowa's new law \"Does [book] contain a description or depiction of a sex act?\"
If ChatGPT said yes, the book was removed from the library.")
More serious:
Just because chatGPT answers questions doesn't mean that it knows anything.
I'm not talking about "it might make a mistake" I'm talking about ... the way it constructs answers isn't based on knowing facts or any logical framework.
The only thing chatGPT cares about is that answers "sound like" they are correct and appropriate in context.
chatGPT is not an oracle, it's not even a search engine, it's not a wikipedia even... it's worse than all these. It's a verbiage engine.
@futurebird
Now what am I going to do with all these typewriters and monkeys ...
@futurebird But... artificial intelligence! End of work. Threat to humanity. Next revolution in productivity. Intelligence breakthrough. Promises were made, dammit. I was expecting more than Eliza 2.0.
@darren @futurebird They might still eventually be fulfilled, but not by anything like these...
@darren @futurebird come on, it is a real threat to people who are basically doing the same bullshit work manually - scrape the internet and re-write the compiled nonsense, mangling it to evade the copyright infringement accusations without any qualified assessment.
@futurebird So, it uses a bunch of dumb tricks to appear smart even though it doesn't have any idea what's going on? Wait... That sounds super familiar...
> verbiage engine
i might need to steal that
to complement what i posted on discord: the power of speech if you completely strip away everything we value speech for
@futurebird somebody referred to it at "mansplaining as a service" and that fits. Full of confidence, light on research and facts.
@futurebird wish more people understood this
@futurebird As Ms. Stein said, "There is no there there" - the only difference it sees between "Men landed on the moon in 1969" and "Men landed on the moon in 1939" is that the latter pattern occurs much less frequently.
@futurebird Yep. A large LANGUAGE model, not a large KNOWLEDGE model. The trouble is that we humans have trained ourselves well over centuries to associate language fluency with intelligence. And the LLMs certainly are fluent, if nothing else.
@futurebird@sauropods.win
It’s a plausibility engine.
@futurebird This is precisely the point I emphasize any time I'm forced to educate people about chatbots and LLMs writ large. They do not "know" or "think" or "believe" or "guess" or "consider" or "assume" or any other verb that implies cognition. They just assemble bits of language together based on patterns they learned from reading what humans have written. Your term "verbiage engine" is my new favorite...thank you for that. 
@futurebird Bullshit generator. It generates text which seems fine, but had no capability to assess the text output in any way.
@futurebird I always tell people that ChatGPT answers are basically created using a set of linked D&D Wandering Monster tables, only for English words.
@futurebird I'm going to get "ChatGPT is glamorised autocomplete, it does not know, want, or think" tattooed on my fist and punch techbros out with it is2g
@futurebird I want to print and frame this and hang a copy next to, I dunno, every computer monitor in the country right now.
@futurebird if I was Open AI, I would go all out against this, fully enforcing their terms of service and issuing a C&D, but we shall see how they play it.
@futurebird "the shape of truth, without truth"
@futurebird
a verbiage engine - great, I will keep that in mind.
@futurebird By definition, Jason Mendoza is smarter than ChatGPT
@futurebird “Verbiage engine” is especially good. I’m also a big fan of “synthetic text extruder,” from @emilymbender
@futurebird What most non-tech people (and a bunch of tech people too) fail to comprehend is that ChatGPT and it’s LLM are in essence super-autocomplete. Hyper rote learning, built on somewhat questionable sampling examples. But yeah, let’s call it AI, because we it’s easy to make people think you’re smart
@futurebird We're just so convinced that AI is intelligent because we've become trained to accept and look for clear sounding reply instead of reasoning. In school, exams and day to day work.
@futurebird I understand fairly well how GPTs work but I like to use them for coding when I'm struggling with the correct syntax. They're just so good at generating something that looks like a duck and even quacks like one. You just throw enough B parameters at it and it will be strikingly convincing in making you think it knows everything.
Years ago my university prof mentioned that with sufficient parameters you can approx any function. I feel this is what's happening now with these GPTs.
@futurebird It's literally the automation of a middle-aged white man whose podcast habit has made him confidently wrong at parties.
@futurebird best description so far! A verbiage engine ... Thank you for the reminder
@futurebird the original office space printer scene
reality: a new technology is developed and the baddies use it for their nefarious plans regardless of any evidence or lack thereof whether it works or not, it doesn't work, it's all bullshit and it doesn't even matter because all the damage is externalized and their only duty was to use it to appease their vibes-based masters
@futurebird well said!
The biggest threat is not that AI knows better than we do, it's that people (us) will accept what is says as being good enough.
I'm sure that's why they market it as fun and 'clever': people are so willing to accept a simple, clearly presented answer.
@futurebird kernigahan's law says it's harder to debug something than to write it. So we'll end up with compound nonsense where people ask gpt, publish it, then gpt ingresses it and spits it out again, like this: