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
hot cocoa on the couch
Dec 8, 2009


lol how easy it was to get this. people that think true ai is close are fuckign rubes

Adbot
ADBOT LOVES YOU

Smugworth
Apr 18, 2003

All of my posts since February 2023 have been made using generative AI. And no one even noticed :awesomelon:

The Moon Monster
Dec 30, 2005

hot cocoa on the couch posted:

lol how easy it was to get this. people that think true ai is close are fuckign rubes

Hey now, we're probably less than a year away from people who have invested heavily in achieving AGI declaring we've achieved AGI.

Saint Isaias Boner
Jan 17, 2007

hi how are you

Smugworth posted:

All of my posts since February 2023 have been made using generative AI. And no one even noticed :awesomelon:

I noticed a huge improvement tbh

Strategic Tea
Sep 1, 2012

Aryan eh, they're not exactly being subtle

Tarkus
Aug 27, 2000

gently caress that's a long prompt. There's no way that any LLM is going to follow all that.

Smugworth
Apr 18, 2003

Saint Isaias Boner posted:

I noticed a huge improvement tbh

:mad:

Tijuana-A-Go-Go
Aug 2, 2019

Doggles Aficionado


Tarkus posted:

gently caress that's a long prompt. There's no way that any LLM is going to follow all that.

but they said "please"

toggle
Nov 7, 2005

I just got paid 41 cents on Pond5 for some stock footage, or in this case “datasets” that “support companies building computer vision and Large Language Models” and to “provide a potential new source of revenue by expanding our reach to new customers and industries, including AI researchers, and leading technology developers and manufacturers”

Wonderful.

Smugworth
Apr 18, 2003

toggle posted:

I just got paid 41 cents on Pond5 for some stock footage, or in this case “datasets” that “support companies building computer vision and Large Language Models” and to “provide a potential new source of revenue by expanding our reach to new customers and industries, including AI researchers, and leading technology developers and manufacturers”

Wonderful.

In the future this will be the only form of work humans do

zedprime
Jun 9, 2007

yospos

Tarkus posted:

gently caress that's a long prompt. There's no way that any LLM is going to follow all that.
They can and do and it's exactly how you make personas out of something that stole half the internet.

It's the weakest form of directing outputs so it's easily broken out of but its capable of following all of those till it isn't.

Duck and Cover
Apr 6, 2007

Smugworth posted:

In the future this will be the only form of work humans do

I think they'll still be opportunity from humilating/dangerous game shows.

Tarkus
Aug 27, 2000

zedprime posted:

They can and do and it's exactly how you make personas out of something that stole half the internet.

It's the weakest form of directing outputs so it's easily broken out of but its capable of following all of those till it isn't.

So in other words, there's no way that any LLM is going to follow all that?

zedprime
Jun 9, 2007

yospos

Tarkus posted:

So in other words, there's no way that any LLM is going to follow all that?
No.

Let me put it another way. One of the ways people have been able to get perceived higher quality responses is by giving models a base charge of conflicting jibber jabber like "think step by step and take your time. Do not think too long. Phrase your response like you are talking to a parakeet." These things love structure, rules, and context. The problem becomes when you give it even more context to your own ends.

Tarkus
Aug 27, 2000

zedprime posted:

No.

Let me put it another way. One of the ways people have been able to get perceived higher quality responses is by giving models a base charge of conflicting jibber jabber like "think step by step and take your time. Do not think too long. Phrase your response like you are talking to a parakeet." These things love structure, rules, and context. The problem becomes when you give it even more context to your own ends.

I honestly don't know what we're arguing about here, perhaps we're talking about different things. You're right that providing structured prompts with context and rules can help steer LLMs to produce more coherent and high-quality outputs to some degree. However, my original point was that extremely long and convoluted prompts with many instructions become unwieldy for even high end language models (They're probably using something like Mistral Large).

There's like 900 tokens, 40+ instructions in that poo poo heap of a pre-prompt. Some people call it cognitive overload or attention overload, whatever the case may be that is probably what is occurring with their model and it is failing to adhere to the instructions it is given. This is why I say "There's no way that any LLM is going to follow all that."

syntaxfunction
Oct 27, 2010
In the most rudimentary version: ya ever notice that even tho you tell it to make a gal.with a blue hat it keeps forgetting either the blue, the hat or the gal? Now imagine that over a mini-dissertation in racism.

mazzi Chart Czar
Sep 24, 2005

syntaxfunction posted:

In the most rudimentary version: ya ever notice that even tho you tell it to make a gal.with a blue hat it keeps forgetting either the blue, the hat or the gal? Now imagine that over a mini-dissertation in racism.

Hitler promised to rid to the world of blue hat people because their trans rights were too leftist.

Strategic Tea
Sep 1, 2012

Smugworth posted:

In the future this will be the only form of work humans do

Everyone should read sci fi heist/caper The Quantum Thief, about a posthuman solar system in which ordinary humans are barely tolerated because their minds can be stripmined by one of the ~12 techbro upload gods for new ideas

naem
May 29, 2011

https://youtube.com/shorts/V7pVnjPvPkg?si=_EmiDyzkyTON9RT5

the AI centipede

Blitter
Mar 16, 2011


I think you need to train your posting LLM with the word "ouroboros".

hellotoothpaste
Dec 21, 2006

I dare you to call it a perm again..

Sekenr posted:

I glued my balls to my butt again is better IMO

Obscurest Vinyls is now my favorite execution of generative AI. Bar none forever.

SidneyIsTheKiller
Jul 16, 2019

I did fall asleep reading a particularly erotic chapter
in my grandmother's journal.

She wrote very detailed descriptions of her experiences...

Failson posted:

Dunno if it's been posted yet, but worth repeating:

https://www.wheresyoured.at/bubble-trouble/

LLM-style AI is already at it's peak usefulness. All the information has been consumed, and the output is already getting worse.

Everything from here is based on vague promises of 'synthetic data' and computing breakthroughs that will not exist.

Just fyi that blog post was generated by google gemini

Snowglobe of Doom
Mar 30, 2012

sucks to be right
Crossposting from the PYF thread, there's a Facebook accoutn that just posts thousands upon thousands of awful AI images of Jesus which keeps getting weirder and weirder

https://twitter.com/CleanestDiesel/status/1778938288897745150

"Squint and you'll see Jesus" is their new gimmick, they've posted thousands of these lovely AI images over the last few weeks and most of them are nightmare fuel
https://www.facebook.com/davon999/photos_by

Random samples:



Their other big obsessions lately has been "Muddy Jesus saves beautiful cabin crew from vehicle" which often randomly features tigers or sharks or lions or eagles and their AI app clearly struggled with this one, but they still kept posting through it


hellotoothpaste
Dec 21, 2006

I dare you to call it a perm again..

Jesus appears, to in fact, be a co-pilot

Edit: :-( the Copilot joke was right there.

hellotoothpaste
Dec 21, 2006

I dare you to call it a perm again..

Snowglobe of Doom posted:

Crossposting from the PYF thread, there's a Facebook accoutn that just posts thousands upon thousands of awful AI images of Jesus which keeps getting weirder and weirder

https://twitter.com/CleanestDiesel/status/1778938288897745150

"Squint and you'll see Jesus" is their new gimmick, they've posted thousands of these lovely AI images over the last few weeks and most of them are nightmare fuel
https://www.facebook.com/davon999/photos_by

Random samples:


Damnit, I was like “this one is a total miss” and then I squinted. It’s an off-center bust shot of Jesus, that’s terrifying

Klyith
Aug 3, 2007

GBS Pledge Week

Tropico 8: This Time Jesus is El Presdente

toggle
Nov 7, 2005


i wonder what the total co2 emissions are for all of these jesus images. or the electricity usage. or water usage.

ultrafilter
Aug 23, 2007

It's okay if you have any questions.



Nobody fucks with the Jesus.

mazzi Chart Czar
Sep 24, 2005

ultrafilter posted:

Nobody fucks with the Jesus.

jesus teams up with john wick in john wick 5

Smugworth
Apr 18, 2003


Nice flesh golem

Tarkus
Aug 27, 2000

Smugworth posted:

Nice flesh golem

Beautiful nussy bottom left.

Tarkus
Aug 27, 2000

Here, have some Trump nussy... kind of.

Dokapon Findom
Dec 5, 2022

They hated Futanari because His posts were shit.
Thanks, love all that garbage, please keep posting it :geno:

Commie Lasorda
May 15, 2009

IT'S CLOBBERIN' TIME!
https://twitter.com/AngstHarrier/status/1779424815335735435

Dokapon Findom
Dec 5, 2022

They hated Futanari because His posts were shit.
We must return to the era of two Caspian Seas :hai:

https://twitter.com/Openatic/status/1779564934185336859

You know those Qatars that are like... Double Qatars, you know?

Impossibly Perfect Sphere
Nov 6, 2002

They wasted Luanne on Lucky!

She could of have been so much more but the writers just didn't care!
Best plan so far.

Impossibly Perfect Sphere
Nov 6, 2002

They wasted Luanne on Lucky!

She could of have been so much more but the writers just didn't care!

quote:

Elon Musk: AI will be smarter than any human around the end of next year

Smarter than him probably.

Vincent Van Goatse
Nov 8, 2006

Enjoy every sandwich.

Smellrose
OP your posts are the only thing in the world worse than AI-generated garbage.

AI at least has an excuse for being lovely and worthless. You could self-improve but you never do.

cumpantry
Dec 18, 2020

woa

Adbot
ADBOT LOVES YOU

hot cocoa on the couch
Dec 8, 2009

Vincent Van Goatse posted:

OP your posts are the only thing in the world worse than AI-generated garbage.

AI at least has an excuse for being lovely and worthless. You could self-improve but you never do.

i don't think he's read tjis thread for over a month

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