Register a SA Forums Account here!
JOINING THE SA FORUMS WILL REMOVE THIS BIG AD, THE ANNOYING UNDERLINED ADS, AND STUPID INTERSTITIAL ADS!!!

You can: log in, read the tech support FAQ, or request your lost password. This dumb message (and those ads) will appear on every screen until you register! Get rid of this crap by registering your own SA Forums Account and joining roughly 150,000 Goons, for the one-time price of $9.95! We charge money because it costs us money per month for bills, and since we don't believe in showing ads to our users, we try to make the money back through forum registrations.
 
  • Post
  • Reply
Roosevelt
Jul 18, 2009

I'm looking for the man who shot my paw.


is that what i think it is? it's beautiful

Adbot
ADBOT LOVES YOU

NoneMoreNegative
Jul 20, 2000
GOTH FASCISTIC
PAIN
MASTER




shit wizard dad

Roosevelt
Jul 18, 2009

I'm looking for the man who shot my paw.


it was doing so well until it got to Coconut

Sagebrush
Feb 26, 2012


this is what happens when you don't give sam altman the 100 trillion dollars he asked for

Beeftweeter
Jun 28, 2005

OFFICIAL #1 GNOME FAN

Sagebrush posted:

this is what happens when you don't give sam altman the 100 trillion dollars he asked for

yeah, at least then it would only get at most 3 out of the 5 wrong, with the remainder being pet treats

fart simpson
Jul 2, 2005

DEATH TO AMERICA
:xickos:


people are clowning on "AI" here but i think what this really shows is how english is a uniquely flexible language

akadajet
Sep 14, 2003

https://www.youtube.com/watch?v=MirzFk_DSiI
I don't know why they decided to make these things sound like they are an actor on one of those shows that pause for a laughtrack every few seconds.

NoneMoreNegative
Jul 20, 2000
GOTH FASCISTIC
PAIN
MASTER




shit wizard dad

https://x.com/ActualNames1/status/1791516799042773439

NoneMoreNegative
Jul 20, 2000
GOTH FASCISTIC
PAIN
MASTER




shit wizard dad




:hmmyes:

Beeftweeter
Jun 28, 2005

OFFICIAL #1 GNOME FAN

ah crap it picked from the "list of ferengi names" database again

fart simpson
Jul 2, 2005

DEATH TO AMERICA
:xickos:

using chat gpt to generate ffmpeg commands for me so i dont have to learn the arcane ffmpeg cli

shitface
Nov 23, 2006

so i had some gently caress around time today and I decided to take a look at gpt 4o. the demo videos were very impressive with the conversational style, the natural sounding conversations and chuckles from the “ai”. well, it doesn’t do any of that. “coming soon”, apparently

anyway, downloaded on iPhone, iPad and Mac. and I ran three experiments

Expiriment 1
the first thing I did was to tell it that i have iPhone, iPad, and Mac and asked it how would i go about downloading the OpenAI gpt client (which I’d just done). it told me I could download the ios clients from the App Store.

I reminded it I wanted a Mac app and it told me I could get it from the Mac App Store (you cannot. it’s not there but a bunch of poo poo trying to look like it’s official sure is). I told it the app isn’t available from the Mac App Store and it apologized but didn’t offer any alternative. then I told it the Mac app is available from the OpenAI website and it claimed to know this all along. I tried to get it to admit its mistake, and first it denied it, then it just threw a “servers are busy bro” error. I’d call this a complete failure.

Experiment 2
I had a spreadsheet with CVEs clearly listed between two version of software. top line vX and vX+1. next line critical, high, medium, low. rows are components. I asked it, “what’s the delta between vX and vX+1?”. I have ABSOLUTELY NO loving IDEA what the numbers it came up with came from. I spent some time seeing if I could see the pattern, but nope. It confidently proclaimed a whole load of bullshit. so, I told it, “no. that’s wrong. for example the delta between vX and vX for component Z is 3”. it said it understood and proceeded to entirely gently caress up table rendering. i asked it to repeat and lo and behold, it rendered the table correctly and got the figures right. so, ok, if you told teenage me you could explain something like this to a computer in plain english and it could appear to introspect and correct its answer, I’d have been impressed. I’ll call this a partial success

Experiment 3
A friend attended an event recently and I wanted to transform a photo she sent me and make it Pixar style. there were five people in the original picture. it produced a (admittedly correctly styled) picture with six people, in poses that did not resemble the original. i told it, “that’s 6 people and there are 5 in the original picture”. it apologized and said “here’s 5 people”… there were 7 in this version. i tried to correct it a few more times and it continually produced between 6 and all the way to 8, each time swearing there were 5. i had to get REALLY specific and commanding before it managed to produce a picture with 5, which it then did reliably after that. it took a whole bunch of more loving around to get it produce something even vaguely resembling the actual poses from the original picture and it had clearly focused on only one person with any accuracy. decent image in the end and my friend enjoyed it but it was not accurate, not what i asked for and, boy, was it a loving pain in the balls to get there. not total failure, but like 1/10th success at best

i ran a bunch of other experiments, some of which made me WTF and others were impressive, coincidentally or not.

i chatted with my boss later, who has some interest in this tech, and told him the same thing as I’ve been telling him for some time. if you can, either by human or by algorithm, validate the output of this thing in significantly less time than the time it would have taken to do the the actual thing, it might be useful for some tasks. otherwise, it’s alphabetti spaghetti with pretensions

shitface fucked around with this message at 15:39 on May 20, 2024

Sagebrush
Feb 26, 2012

Thanks for testing it out for us, shitface.

shitface
Nov 23, 2006

Sagebrush posted:

Thanks for testing it out for us, shitface.

I’m assuming sarcasm here

in any case, i don’t think this poo poo is worth boiling the ocean for rn. you could do most of it what it accomplishes by classical algorithms and the other stuff, well, you could always hire some people

Sagebrush
Feb 26, 2012

No sarcasm, I just wanted to respond that way with your username

shitface
Nov 23, 2006

Sagebrush posted:

No sarcasm, I just wanted to respond that way with your username

“thanks forums user shitface” would have been the more conventional approach imo

Dans Macabre
Apr 24, 2004



This is apparently this guy Chafic's fault. https://www.quora.com/What-fruit-name-ends-with-UM

I'm emailing him to learn more.

Arcsech
Aug 5, 2008
ime for text stuff gpt 4t/o are roughly equivalent to a reasonably intelligent but extremely stonerbrained college student intern. reasonably capable of following thorough directions, will absolutely not hesitate to confidently bullshit or gaslight you about anything and everything, attention span of a dead goldfish, will randomly just start doing what it was taught in class instead of what you asked or anything that makes sense if it gets confused.

like, a stonerbrained intern can be reasonably valuable. but uh, it might be slightly overblown in terms of ability

Roosevelt
Jul 18, 2009

I'm looking for the man who shot my paw.

shitface posted:

so i had some gently caress around time today and I decided to take a look at gpt 4o. the demo videos were very impressive with the conversational style, the natural sounding conversations and chuckles from the “ai”. well, it doesn’t do any of that. “coming soon”, apparently

if my computer ever “chuckles” at me i will torture and kill it

also, thanks shitface

Juul-Whip
Mar 10, 2008

shitface posted:

so i had some gently caress around time today and I decided to take a look at gpt 4o.
It confidently proclaimed a whole load of bullshit.

wow i can't believe it

Beeftweeter
Jun 28, 2005

OFFICIAL #1 GNOME FAN

fart simpson posted:

using chat gpt to generate ffmpeg commands for me so i dont have to learn the arcane ffmpeg cli

i forget which thread it was in but maybe a year ago someone asked me how to do some ffmpeg thing, and i replied with the correct invocation

but then they said they wanted to understand how i did that better, so they pasted it into some site that used ai where you could paste the command in, it'd look at manpages, then explain what the switches do, which sounded cool and potentially helpful even

but they said it couldn't determine what most of i was using did. apparently one reason was because i had used their short form when available, but that shouldn't have mattered, since i know the manpage definitely has them listed

eventually i figured out the source manpage was some ancient version by trying some deprecated features, apparently it just didn't have some of the switches because it was too old lol. but either way most of the shortened forms were still valid syntax back then :shrug:

akadajet
Sep 14, 2003

shitface posted:

so i had some gently caress around time today and I decided to take a look at gpt 4o. the demo videos were very impressive with the conversational style, the natural sounding conversations and chuckles from the “ai”. well, it doesn’t do any of that. “coming soon”, apparently

anyway, downloaded on iPhone, iPad and Mac. and I ran three experiments

Expiriment 1
the first thing I did was to tell it that i have iPhone, iPad, and Mac and asked it how would i go about downloading the OpenAI gpt client (which I’d just done). it told me I could download the ios clients from the App Store.

I reminded it I wanted a Mac app and it told me I could get it from the Mac App Store (you cannot. it’s not there but a bunch of poo poo trying to look like it’s official sure is). I told it the app isn’t available from the Mac App Store and it apologized but didn’t offer any alternative. then I told it the Mac app is available from the OpenAI website and it claimed to know this all along. I tried to get it to admit its mistake, and first it denied it, then it just threw a “servers are busy bro” error. I’d call this a complete failure.

Experiment 2
I had a spreadsheet with CVEs clearly listed between two version of software. top line vX and vX+1. next line critical, high, medium, low. rows are components. I asked it, “what’s the delta between vX and vX+1?”. I have ABSOLUTELY NO loving IDEA what the numbers it came up with came from. I spent some time seeing if I could see the pattern, but nope. It confidently proclaimed a whole load of bullshit. so, I told it, “no. that’s wrong. for example the delta between vX and vX for component Z is 3”. it said it understood and proceeded to entirely gently caress up table rendering. i asked it to repeat and lo and behold, it rendered the table correctly and got the figures right. so, ok, if you told teenage me you could explain something like this to a computer in plain english and it could appear to introspect and correct its answer, I’d have been impressed. I’ll call this a partial success

Experiment 3
A friend attended an event recently and I wanted to transform a photo she sent me and make it Pixar style. there were five people in the original picture. it produced a (admittedly correctly styled) picture with six people, in poses that did not resemble the original. i told it, “that’s 6 people and there are 5 in the original picture”. it apologized and said “here’s 5 people”… there were 7 in this version. i tried to correct it a few more times and it continually produced between 6 and all the way to 8, each time swearing there were 5. i had to get REALLY specific and commanding before it managed to produce a picture with 5, which it then did reliably after that. it took a whole bunch of more loving around to get it produce something even vaguely resembling the actual poses from the original picture and it had clearly focused on only one person with any accuracy. decent image in the end and my friend enjoyed it but it was not accurate, not what i asked for and, boy, was it a loving pain in the balls to get there. not total failure, but like 1/10th success at best

i ran a bunch of other experiments, some of which made me WTF and others were impressive, coincidentally or not.

i chatted with my boss later, who has some interest in this tech, and told him the same thing as I’ve been telling him for some time. if you can, either by human or by algorithm, validate the output of this thing in significantly less time than the time it would have taken to do the the actual thing, it might be useful for some tasks. otherwise, it’s alphabetti spaghetti with pretensions

did it tell you all that in that horny female voice?

akadajet
Sep 14, 2003

oh, re-read the beginning of your post. no horny voice yet

Agile Vector
May 21, 2007

scrum bored



Arcsech posted:

ime for text stuff gpt 4t/o are roughly equivalent to a reasonably intelligent but extremely stonerbrained college student intern. reasonably capable of following thorough directions, will absolutely not hesitate to confidently bullshit or gaslight you about anything and everything, attention span of a dead goldfish, will randomly just start doing what it was taught in class instead of what you asked or anything that makes sense if it gets confused.

like, a stonerbrained intern can be reasonably valuable. but uh, it might be slightly overblown in terms of ability

this maps similarly to my mental model of siri and other phone assistants, which is equivalent to a new hire that got their job because they're their uncle's favorite. kinda good, sometimes on the ball, but not consistently enough to be reliable

Agile Vector
May 21, 2007

scrum bored



akadajet posted:

did it tell you all that in that horny female voice?

openai removed the scarjo voice today after getting called out for it

Luigi Thirty
Apr 30, 2006

Emergency confection port.

Agile Vector posted:

openai removed the scarjo voice today after getting called out for it

scarjo statement dropped via npr and lmao sam altman pitched her on voicing chatgpt, she said no, and they just used her voice from a movie anyway

https://x.com/BobbyAllyn/status/1792679435701014908

shitface
Nov 23, 2006

dammit, i had picked the horny scarjo voice too 😟

kliras
Mar 27, 2021
they would have warned us if not for the noncon disclosure agreements

seriously, i wonder what the final straw was for the people who left openai

ziasquinn
Jan 1, 2006

Fallen Rib

Luigi Thirty posted:

scarjo statement dropped via npr and lmao sam altman pitched her on voicing chatgpt, she said no, and they just used her voice from a movie anyway

https://x.com/BobbyAllyn/status/1792679435701014908

so great you could interrogate it about sounding just like her and it would just say some mealy mouthed poo poo about it just being aggregated ok sure

Beeftweeter
Jun 28, 2005

OFFICIAL #1 GNOME FAN
don't worry i'm sure they just fed it every line she's ever said in any recording they were able to find

notwithoutmyanus
Mar 17, 2009

Beeftweeter posted:

don't worry i'm sure they just fed it every line she's ever said in any recording they were able to find

Seems openAI has the same concept of consent as crypto: none

Beeftweeter
Jun 28, 2005

OFFICIAL #1 GNOME FAN

notwithoutmyanus posted:

Seems openAI has the same concept of consent as crypto: none

they very obviously don't care about intellectual property in general, unless it's theirs, in which case scraping it would make you literally worse than hitler

fart simpson
Jul 2, 2005

DEATH TO AMERICA
:xickos:

Beeftweeter posted:

they very obviously don't care about intellectual property in general, unless it's theirs, in which case scraping it would make you literally worse than hitler

i think theyve learned their lesson and will care about it going forward op

~Coxy
Dec 9, 2003

R.I.P. Inter-OS Sass - b.2000AD d.2003AD

akadajet posted:

did it tell you all that in that horny female voice?

ChatGPT has finally caught up to Unreal Tournament 2004

edit: guess I really should have read the whole page before replying, huh

~Coxy fucked around with this message at 04:47 on May 21, 2024

akadajet
Sep 14, 2003

Agile Vector posted:

openai removed the scarjo voice today after getting called out for it

I didn’t know that’s what they were going for.

Agile Vector
May 21, 2007

scrum bored



chatgpt : intelligence :: ut2004 physics engine : realism

Agile Vector
May 21, 2007

scrum bored



akadajet posted:

I didn’t know that’s what they were going for.

I think sam tried to play it off as a sound-alike performer (and maybe that's true) as a fig leaf but tweeting out "her" right after is a footgun if ever there was one

I didn't hear the announcement our try it so idk if it sounds like her voice :shrug:

edit: the trend of incredibly sad rich losers tweeting one-word poo poo that makes them look desperate is so dumb

akadajet
Sep 14, 2003

Agile Vector posted:

I think sam tried to play it off as a sound-alike performer (and maybe that's true) as a fig leaf but tweeting out "her" right after is a footgun if ever there was one

I didn't hear the announcement our try it so idk if it sounds like her voice :shrug:

it sounds like a computer scripted to sound really flirty. I never watched her. it looks cringy and I dislike the color grading in the trailer

Carthag Tuek
Oct 15, 2005

Tider skal komme,
tider skal henrulle,
slægt skal følge slægters gang



akadajet posted:

it sounds like a computer scripted to sound really flirty. I never watched her. it looks cringy and I dislike the color grading in the trailer

its a good movie

Adbot
ADBOT LOVES YOU

fart simpson
Jul 2, 2005

DEATH TO AMERICA
:xickos:

is this the end of authorship?

https://github.com/Bklieger/groqbook

  • 1
  • 2
  • 3
  • 4
  • 5
  • Post
  • Reply