Chat GPT AI

  • Guest, it's time once again for the hotly contested and exciting FoH Asshat Tournament!



    Go here and fill out your bracket!
    Who's been the biggest Asshat in the last year? Once again, only you can decide!

pharmakos

soʞɐɯɹɐɥd
<Bronze Donator>
16,306
-2,239
I wonder if it's looking at your browser cookies or its local storage. Try the same thing from two different computers, or in an incognito window.
It shouldn't have access to that stuff, should it? Basic permissions stuff should block that I'd think.
 

Lambourne

Ahn'Qiraj Raider
2,688
6,487
Also just take the whole thing with a grain of salt, because the group of people who performed these tests are basically a cult who believe in nonsense like Roko's basilisk - Wikipedia, something I have tried many times to understand what they view as profoundly dangerous and come up with nothing.

I'd never heard about this until I read your post so I read up on it. Not 48 hours later on the Midjourney discord:

1679473001754.png


The matrix is glitching. It knows we know.
 

Sanrith Descartes

Veteran of a thousand threadban wars
<Aristocrat╭ರ_•́>
41,351
107,244
Ok, how is this possible? How can copy pasting the exact same math problem getting two different answers. Isn't a typo. I copy/pasted the query. Its math, how can an AI get two different answers using the same formula and inputs?

1679536494581.png
 
  • 1WTF
Reactions: 1 user

Captain Suave

Caesar si viveret, ad remum dareris.
4,678
7,920
Ok, how is this possible? How can copy pasting the exact same math problem getting two different answers. Isn't a typo. I copy/pasted the query. Its math, how can an AI get two different answers using the same formula and inputs?

View attachment 465109

There is some randomness in the responses, and since it's a text prediction machine and not actually doing math this is just two similar but different hallucinations of a math problem's answer.

Edit: The correct answer is 7282.8 ft^3, so not only was it inconsistent but it was wrong twice.
 
Last edited:
  • 1Worf
  • 1Like
Reactions: 1 users

Sanrith Descartes

Veteran of a thousand threadban wars
<Aristocrat╭ರ_•́>
41,351
107,244
There is some randomness in the responses, and since it's a text prediction machine and not actually doing math this is just two similar but different hallucinations of a math problem's answer.

Edit: The correct answer is 7282.8 ft^3, so not only was it inconsistent but it was wrong twice.
I know. I was like wtf? Ask it what the formula is and its good. Ask it to solve the problem and not so much. My guess is they trained it on common core math.
 
  • 2Worf
Reactions: 1 users

Control

Ahn'Qiraj Raider
2,096
5,209
Ok, how is this possible? How can copy pasting the exact same math problem getting two different answers. Isn't a typo. I copy/pasted the query. Its math, how can an AI get two different answers using the same formula and inputs?

View attachment 465109
If it works like the ai image tools, every response is randomized based on a seed. In Midjourney, you can pull/input the seed to get predictable results, not sure if chatai lets you do that. That doesn't say much about the bad math of course, but some rolls probably have better assumptions than others.
 
  • 1Like
Reactions: 1 user

Rajaah

Honorable Member
<Gold Donor>
11,002
14,655
What did the robot say that was woke?

Probably nothing, the Hodge Twins are just pre-emptively afraid of future robot police that are programmed to be woke and arrest people for saying the wrong thing, I guess.

I think this nightmare scenario is a lot more likely:


You just know the military can't resist putting AI in charge of autonomous weapons, and that's a disaster waiting to happen. James Cameron's Terminator movies might have actually made a substantial difference in the future, ironically, because they became so ingrained in pop culture that you can't even bring this kind of tech up in a meeting without people balking at it. They should balk at it.

Now: ChatGPT helps us write thoughtful messages!

Soon: Slaughterbots
 

Sanrith Descartes

Veteran of a thousand threadban wars
<Aristocrat╭ರ_•́>
41,351
107,244
If it works like the ai image tools, every response is randomized based on a seed. In Midjourney, you can pull/input the seed to get predictable results, not sure if chatai lets you do that. That doesn't say much about the bad math of course, but some rolls probably have better assumptions than others.
But there shouldn't be any randomness involved. It can obviously have knowledge of the order of operations in math and it knew the correct formula. It only had to apply a variable in the formula.
 

Control

Ahn'Qiraj Raider
2,096
5,209
But there shouldn't be any randomness involved. It can obviously have knowledge of the order of operations in math and it knew the correct formula. It only had to apply a variable in the formula.
That would make sense of course, but it's not doing math. It's just giving you the words it thinks are most relevant to your question. If you get different answers to the same question, then there's clearly some random seeding going on.
 
  • 2Like
Reactions: 1 users

ShakyJake

<Donor>
7,597
19,181
I know. I was like wtf? Ask it what the formula is and its good. Ask it to solve the problem and not so much. My guess is they trained it on common core math.
There is a setting that forces it to be deterministic, removing any randomness from the responses. I wonder if that would've made a difference?
 
  • 1Like
Reactions: 1 user

Sanrith Descartes

Veteran of a thousand threadban wars
<Aristocrat╭ರ_•́>
41,351
107,244
I just figured this out. Yes I am slow. If GPT is answering a query and stops due to length, you can just say "finish creating the document above beginning at bullet point "x"" and it continues on in the next reply.
 
  • 1Like
Reactions: 1 user

Mist

Eeyore Enthusiast
<Gold Donor>
30,272
22,006
Bard fucking sucks.

1679715494876.png


Vs GPT:

Silly mmo character names

Fuzzy Picklebottom
Sir Puddles McFluffernutter
Professor Wigglesnout
Baron Von Chickenpants
Lady Bubbles the Fierce
Captain SassyPants
Duke Fluffernutter
Princess Sparklebutt
Count Snuggles the Brave
Sir Fuzzypants the Mighty
Lord Wigglytush
Queen Snugglebug
Lady Gigglesworth
Sir Meowington
Count Fluffykins the Terrible
King Cuddlebug
Duchess Fluffernutter
Prince Wiggles the Wobbly
Lady Furrybottom
Captain Cuddlefluff


Silly MMO character names that are a single word

Snugglepuff
Fuzzleboop
Glitterbomb
Puddletop
Wigglesnoot
Fluffernugget
Snickerdoodle
Bumblesnout
Sparklepop
Furrypants
Gigglewig
Fuzzykins
Bouncybottom
Wiggletail
Snugglebuns
Gigglesnort
Fluffysnack
Snugglefluff
Puddlesnoot
Fuzzybump


Less cute

Bloodspike
Deathblade
Skullcrusher
Doombringer
Shadowslayer
Nightstalker
Hellfire
Blackheart
Grimreaper
Demonhunter
Warlock
Necromancer
Darklord
Deathknight
Blackout
Nightshade
Bonecrusher
Darkblade
Deathstrike
Vengeance


Less edgy, more creative

Whimsyweaver
Starlighter
Moonbeam
Bubbleglimmer
Spellbinder
Dreamcatcher
Enchantress
Sunflower
Rainbowrider
Skywatcher
Shadowdancer
Fireflyer
Songbird
Thunderbolt
Seashell
Earthsong
Nightingale
Sunbeam
Rainbowmist
Mysticwater



Less literal

Velvetwhisper
Moonwhisperer
Mysticmuse
Serenitykeeper
Nightbloom
Stardancer
Thunderheart
Shadowdreamer
Crystalmind
Enigmaquest
Dreamjester
Emberwhirl
Stormwatch
Echoesofmagic
Celestialpulse
Silkenspirit
Chromatica
Luminousrealm
Mythicmaze
Illusionist



More like names, less like words

Kaelthas
Tharion
Elendria
Varianth
Eadric
Lyriana
Talindra
Lirien
Faeryl
Aurius
Sylphrena
Eldrathil
Tormund
Altheira
Kethryll
Valyria
Arcturus
Thelania
Baelenor
Myrddin




More original

Zerynthia
Calandria
Elyndria
Neryssa
Aeloria
Sylveria
Valerianth
Kestrellia
Aeristia
Valtorin
Asherind
Faeloria
Sylphera
Eiridian
Lyrelle
Cyndrael
Ithrilor
Ophiria
Velendria
Zalindra



Okay, more like those, but more masculine

Valerius
Zanderth
Thorneus
Dariusson
Kaelron
Dravenor
Rylandar
Theronis
Zarekth
Magnusson
Valtorin
Alaricus
Caelion
Vaelric
Torion
Asherion
Darienar
Zephyrion
Cygnusar
Orionth




Slightly more playful

Fizzwick
Sparken
Zestor
Jaxxen
Waverly
Twiggles
Dashiel
Pippin
Skyley
Rivenn
Brambley
Nixton
Azulen
Kipper
Wystan
Frenton
Cobbley
Zealor
Eldwin
Fyren



Perfect, more like those

Solwyn
Havenwood
Sylphson
Stardale
Emberby
Kaelen
Wilderick
Rivenn
Faelon
Valtair
Thorneby
Jaxxson
Lyrenn
Skyleth
Asherby
Axlion
Zephyrus
Cobblewick
Vesperin
Eldrake
 
  • 4Like
Reactions: 3 users

slippery

<Bronze Donator>
7,886
7,701
I used it to make up some resume shit (something I'm terrible at and hate) and it's pretty damn good for it.
 
  • 1Like
  • 1Genius!
Reactions: 1 users

Aldarion

Egg Nazi
8,812
24,053
Yudkowsky proposes international cooperation, even between rivals like the U.S. and China, to shut down development of powerful AI systems. He says this is more important than "preventing a full nuclear exchange," and that countries should even consider using nuclear weapons "if that's what it takes to reduce the risk of large AI training runs."
I guess its pretty widely accepted that the current generation of LLM projects like chatGPT, are not self-aware beings that pose any kind of threat to humanity. So people may shrug off statements like the above as being crackpots.

But I think its also pretty clear that we are collectively going full speed ahead with this stuff, and the only limits we're really hearing about are "lets make sure it doesnt say things that offend people".

Perhaps there is a middle ground somewhere between "launch nukes to prevent training runs" and "the only limits are don't offend leftists"?
 

slippery

<Bronze Donator>
7,886
7,701

I guess its pretty widely accepted that the current generation of LLM projects like chatGPT, are not self-aware beings that pose any kind of threat to humanity. So people may shrug off statements like the above as being crackpots.

But I think its also pretty clear that we are collectively going full speed ahead with this stuff, and the only limits we're really hearing about are "lets make sure it doesnt say things that offend people".

Perhaps there is a middle ground somewhere between "launch nukes to prevent training runs" and "the only limits are don't offend leftists"?
Even just a year ago I'd have said you where crazy to be worried about AI. Now we could be toast in our lifetime.
 
  • 1Like
Reactions: 1 user