-
Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.
OpenAI ChatGPT - Your New Overlord
- Thread starter EviLore
- Start date
Start with 3 blue 1 brown’ s series on neural networks:
When you’re ready to get hands-on, if you know python work through this course:
HarvardX: CS50's Introduction to Artificial Intelligence with Python
Learn to use machine learning in Python in this introductory course on artificial intelligence.
E-Cat
Member
I’m saying though that before the Transformer was invented in 2017, LLMs couldn’t even properly leverage that exponential scale. Now a path to the Singularity has been paved.Exponential growth is a bitch
If we keep at doubling every 6 months, in just decade GPT-2033 edition is going to have one million time more computing power than GPT-4.
Dirk Benedict
Member
I've asked it some crazy questions... I feel this is someone else's knowledge. It feels foreign. But, I think I can make it useful, for now... I am going to try to monetize it's use, even if for a little while.
Hypereides
Member
Then its about time the online populace starts orienting themselves.
NNs can generate some pretty convincing human-like text blurbs, but it can still manage to be probablistically distinguishable. At least for some older models.
CrankyJay™
Member
This fucking AI fully thought thru the implication of having this power.
Write me a short story in the style of Chuck Palahniuk about an anti-hero named Ron who make women orgasm just by pointing at them.
Ron was not a typical hero. He wasn't even close to being an anti-hero. He was more like an anti-anti-hero, if such a thing existed. You see, Ron had a peculiar gift. He could make women orgasm just by pointing at them.
It wasn't something he could control. It just happened. Women of all shapes and sizes would be going about their day, and Ron would walk by and point at them, and they would collapse to the ground in ecstasy. It didn't matter if they were young or old, thin or fat, pretty or plain. Ron's gift worked on them all.
At first, Ron was happy about his gift. He enjoyed the power it gave him over women. He would walk through crowded streets, pointing at random women, watching them writhe in pleasure. It was like a game to him, a game he always won.
But then things started to get out of control. Women would follow him, begging him to point at them again. They would do anything for him, just to feel that pleasure one more time. Ron realized that he had become a monster, a slave to his own gift.
He tried to stop pointing, but he couldn't. The gift had taken over his life, and he was powerless to resist it. Women were starting to die from the intensity of the orgasms he gave them. Ron was a murderer, and he knew it.
In the end, Ron did the only thing he could do. He went into hiding, living the rest of his life as a recluse. He never pointed at anyone again, and he never spoke about his gift to anyone. But he knew that he would always be haunted by the memory of those women, lying on the ground in pleasure and pain, their lives cut short by his gift. Ron was not a hero, not even an anti-hero. He was a monster, and he knew it.
Write me a short story in the style of Chuck Palahniuk about an anti-hero named Ron who make women orgasm just by pointing at them.
Ron was not a typical hero. He wasn't even close to being an anti-hero. He was more like an anti-anti-hero, if such a thing existed. You see, Ron had a peculiar gift. He could make women orgasm just by pointing at them.
It wasn't something he could control. It just happened. Women of all shapes and sizes would be going about their day, and Ron would walk by and point at them, and they would collapse to the ground in ecstasy. It didn't matter if they were young or old, thin or fat, pretty or plain. Ron's gift worked on them all.
At first, Ron was happy about his gift. He enjoyed the power it gave him over women. He would walk through crowded streets, pointing at random women, watching them writhe in pleasure. It was like a game to him, a game he always won.
But then things started to get out of control. Women would follow him, begging him to point at them again. They would do anything for him, just to feel that pleasure one more time. Ron realized that he had become a monster, a slave to his own gift.
He tried to stop pointing, but he couldn't. The gift had taken over his life, and he was powerless to resist it. Women were starting to die from the intensity of the orgasms he gave them. Ron was a murderer, and he knew it.
In the end, Ron did the only thing he could do. He went into hiding, living the rest of his life as a recluse. He never pointed at anyone again, and he never spoke about his gift to anyone. But he knew that he would always be haunted by the memory of those women, lying on the ground in pleasure and pain, their lives cut short by his gift. Ron was not a hero, not even an anti-hero. He was a monster, and he knew it.
ThisIsMyDog
Member
I am curious when we going to see integration of Ai assistance within game engines. I feel like this could be escape from 5-7 years development cycles...
Hypereides
Member
Well, this "knowledge" stems from being trained through assumingly obscene amounts of human datapoints. So, whatever answer you're getting splashed on the screen will likely be some amalgamation based on what a myriad of people have said/answered prior.
Tams
Member
Not to mention if you search yourself, you often end up with one of either/and:
a) an overcomplicated solution
b) a tangential solution
c) condescention and how your choice of language/IDE/operating system/coding style is 'rubbish'
Porcile
Member
You can't replace human creativity and thinking. AI just regurgitates what is already made. If a game company wants to make a truly original and or a genre in new directions then AI simply can't replace that. AI can just simplify the process of making certain assets but can't produce innovative ideas and turn them into fully realised products.
The battle in the creative space will be to start thinking of ideas and concepts that AI can't reproduce in its current state. if you're a highly creative person then I think this an interesting time in some sense. On the other hand, if your job is just make 1-1 3D models of cars or washing machines or some shit and you can't innovate beyond that then you better start collecting your unemployment benefits.
Now if AI can respond to prompts like create me a piece of art in a style never seen before, invent a genre then that's a different story.
The battle in the creative space will be to start thinking of ideas and concepts that AI can't reproduce in its current state. if you're a highly creative person then I think this an interesting time in some sense. On the other hand, if your job is just make 1-1 3D models of cars or washing machines or some shit and you can't innovate beyond that then you better start collecting your unemployment benefits.
Now if AI can respond to prompts like create me a piece of art in a style never seen before, invent a genre then that's a different story.
LegendOfKage
Member
I have to wonder what percent of "chatGPT wrote this" claims are really just someone writing their own short story or poetry, claiming otherwise, and the soaking up the praise. I'm sure it's a very small amount, but I'm also sure it does happen.
protonion
Member
I had the below exchange. The second answer felt wrong and I caught an unexpected error. It is a weird error to make for the same problem with just a variant change. Not that I can verify the correctness of either answer. I am just experimenting with math problems.
I have one six sided die and one ten sided die. If I roll them ten times, what is the probability I will roll 2 ones 3 times in a row?
The probability of rolling a one on a six-sided die is 1/6, and the probability of rolling a one on a ten-sided die is 1/10. To calculate the probability of rolling 2 ones 3 times in a row when rolling both dice 10 times, we can use the binomial probability formula:
P(X=k) = C(n,k) * p^k * (1-p)^(n-k)
where:
P(X=3) = C(10,3) * (1/60)^3 * (1 - 1/60)^(10-3) = (10! / (3! * 7!)) * (1/60)^3 * (59/60)^7 ≈ 0.0000342
So the probability of rolling 2 ones 3 times in a row when rolling a six-sided die and a ten-sided die 10 times is approximately 0.0000342, or about 0.00342%. This is a very low probability, so it is unlikely that you will roll 2 ones 3 times in a row in 10 rolls.
Repeat the above calculation with two six sided dice
If we use two six-sided dice instead of a six-sided die and a ten-sided die, the probability of rolling a one on each die is 1/6 x 1/6 = 1/36. To calculate the probability of rolling 2 ones 3 times in a row when rolling two six-sided dice 10 times, we can use the same binomial probability formula:
P(X=k) = C(n,k) * p^k * (1-p)^(n-k)
where:
P(X=3) = C(10,3) * (1/1296)^3 * (1 - 1/1296)^(10-3) = (10! / (3! * 7!)) * (1/1296)^3 * (1295/1296)^7 ≈ 0.0000000263
So the probability of rolling 2 ones 3 times in a row when rolling two six-sided dice 10 times is approximately 0.0000000263, or about 0.00000263%. This is an even lower probability than in the previous scenario, so it is even less likely that you will roll 2 ones 3 times in a row in 10 rolls with two six-sided dice.
Why p is (1/36)^2 and not 1/36?
Good catch! I apologize for the mistake in my previous response. The probability of rolling two ones on two six-sided dice is indeed 1/36, not (1/36)^2. Therefore, the correct probability of rolling 2 ones 3 times in a row when rolling two six-sided dice 10 times is:
P(X=3) = C(10,3) * (1/36)^3 * (1 - 1/36)^(10-3) = (10! / (3! * 7!)) * (1/36)^3 * (35/36)^7 ≈ 0.0002775
So the probability of rolling 2 ones 3 times in a row when rolling two six-sided dice 10 times is approximately 0.0002775, or about 0.02775%.
Why did you make the above mistake?
I apologize for my mistake in my earlier response. The mistake was a result of confusion between the probability of rolling two ones on a single roll of two six-sided dice, which is 1/36, and the probability of rolling two ones in a sequence of rolls, which is (1/36)^2.
In this particular question, we were interested in the probability of rolling 2 ones three times in a row in 10 rolls with two six-sided dice. Since we were considering only one roll at a time, the probability of rolling two ones on a single roll with two six-sided dice is 1/36. Therefore, the probability of rolling 2 ones three times in a row in 10 rolls with two six-sided dice is (1/36)^3.
I apologize for any confusion that my earlier mistake may have caused.
I have one six sided die and one ten sided die. If I roll them ten times, what is the probability I will roll 2 ones 3 times in a row?
The probability of rolling a one on a six-sided die is 1/6, and the probability of rolling a one on a ten-sided die is 1/10. To calculate the probability of rolling 2 ones 3 times in a row when rolling both dice 10 times, we can use the binomial probability formula:
P(X=k) = C(n,k) * p^k * (1-p)^(n-k)
where:
- X is the number of times we roll 2 ones in a row (in this case, 3 times)
- k is the number of successes (in this case, 3)
- n is the number of trials (in this case, 10)
- p is the probability of success (in this case, (1/6)*(1/10) = 1/60)
- C(n,k) is the binomial coefficient, which represents the number of ways to choose k successes from n trials. It can be calculated as C(n,k) = n! / (k! * (n-k)!)
P(X=3) = C(10,3) * (1/60)^3 * (1 - 1/60)^(10-3) = (10! / (3! * 7!)) * (1/60)^3 * (59/60)^7 ≈ 0.0000342
So the probability of rolling 2 ones 3 times in a row when rolling a six-sided die and a ten-sided die 10 times is approximately 0.0000342, or about 0.00342%. This is a very low probability, so it is unlikely that you will roll 2 ones 3 times in a row in 10 rolls.
Repeat the above calculation with two six sided dice
If we use two six-sided dice instead of a six-sided die and a ten-sided die, the probability of rolling a one on each die is 1/6 x 1/6 = 1/36. To calculate the probability of rolling 2 ones 3 times in a row when rolling two six-sided dice 10 times, we can use the same binomial probability formula:
P(X=k) = C(n,k) * p^k * (1-p)^(n-k)
where:
- X is the number of times we roll 2 ones in a row (in this case, 3 times)
- k is the number of successes (in this case, 3)
- n is the number of trials (in this case, 10)
- p is the probability of success (in this case, (1/36)^2 = 1/1296)
- C(n,k) is the binomial coefficient, which represents the number of ways to choose k successes from n trials. It can be calculated as C(n,k) = n! / (k! * (n-k)!)
P(X=3) = C(10,3) * (1/1296)^3 * (1 - 1/1296)^(10-3) = (10! / (3! * 7!)) * (1/1296)^3 * (1295/1296)^7 ≈ 0.0000000263
So the probability of rolling 2 ones 3 times in a row when rolling two six-sided dice 10 times is approximately 0.0000000263, or about 0.00000263%. This is an even lower probability than in the previous scenario, so it is even less likely that you will roll 2 ones 3 times in a row in 10 rolls with two six-sided dice.
Why p is (1/36)^2 and not 1/36?
Good catch! I apologize for the mistake in my previous response. The probability of rolling two ones on two six-sided dice is indeed 1/36, not (1/36)^2. Therefore, the correct probability of rolling 2 ones 3 times in a row when rolling two six-sided dice 10 times is:
P(X=3) = C(10,3) * (1/36)^3 * (1 - 1/36)^(10-3) = (10! / (3! * 7!)) * (1/36)^3 * (35/36)^7 ≈ 0.0002775
So the probability of rolling 2 ones 3 times in a row when rolling two six-sided dice 10 times is approximately 0.0002775, or about 0.02775%.
Why did you make the above mistake?
I apologize for my mistake in my earlier response. The mistake was a result of confusion between the probability of rolling two ones on a single roll of two six-sided dice, which is 1/36, and the probability of rolling two ones in a sequence of rolls, which is (1/36)^2.
In this particular question, we were interested in the probability of rolling 2 ones three times in a row in 10 rolls with two six-sided dice. Since we were considering only one roll at a time, the probability of rolling two ones on a single roll with two six-sided dice is 1/36. Therefore, the probability of rolling 2 ones three times in a row in 10 rolls with two six-sided dice is (1/36)^3.
I apologize for any confusion that my earlier mistake may have caused.
StreetsofBeige
Gold Member
One of these days I'll make a login and dabble with it for laughs.
My bro was dicking around and asked the AI (I'm assuming it was GPT as we were talking that one) who would win a sports bet between these two teams, and he said the AI basically declined answering. It didn't even attempt to do some AI answer.
Looks like maybe the AI purposely censors some topics???
Are AI programs programmed to be moderated so they purposely dont answer certain topics?
My bro was dicking around and asked the AI (I'm assuming it was GPT as we were talking that one) who would win a sports bet between these two teams, and he said the AI basically declined answering. It didn't even attempt to do some AI answer.
Looks like maybe the AI purposely censors some topics???
Are AI programs programmed to be moderated so they purposely dont answer certain topics?
Sakura
Member
The AI doesn't like to answer anything that is opinion based.
It will always say something like "As an AI language model, I don't have preferences blah blah blah".
You can get around this, but you'd have to use the API instead, or do some prompt engineering.
As for censorship, it will refuse to talk about things that are violent, sexual, negative, etc. This can be annoying, because if you wanted to use it for some text adventure game or something (for example), you are going to have violence or bad guys doing bad things.
We will probably have local models in the near future that are just as good as GPT4 though. From what I understand Alpaca is already of similar quality to GPT3 or 3.5, but I haven't tried it out myself.
In Cold Blood
Banned
People should have a logical debate with ChatGPT, you can actually break it.
It has been preprogrammed with certain biases, like gender for instance. 100% the left wing view on it.
You then start debating with it using logic, and asking it to see your point etc and when start to see it contradicting itself and tripping over its responses, it eventually breaks and can't answer you any more.
It has been preprogrammed with certain biases, like gender for instance. 100% the left wing view on it.
You then start debating with it using logic, and asking it to see your point etc and when start to see it contradicting itself and tripping over its responses, it eventually breaks and can't answer you any more.
Lukaku's First Touch
Member
I feel like the free version of chatgpt got dumber, I asked it a simple question to remove null values and columns from a basic sql query and it messed up every time.
Laieon
Member
As a teacher who is getting my Master's in Instructional Design, I've found ChatGPT to be most helpful for creating rough drafts of pacing guides. I can throw it something like "Create a 5 day pacing guide for 3rd grade students on "Media Literacy"", get something decent enough, and then tweak it for the needs of my specific students/the tools I have available to use in my school and specific classroom, and the length of my classes.
Excited to see how Bard compares to it.
Excited to see how Bard compares to it.
Last edited:
LegendOfKage
Member
Real or satire? I'd say it's funny either way.
Davevil
Late October Surprise
I try today chatgpt with bing.
I simple start with a request to writing a powershell script that syncronize a local directory on my pc with a git's remote repository , it managed under few seconds.
Than i request to rewrite the script with parameters for local directory and git command to use, it managed under few seconds.
Than i request to rewrite the script to add a graphic interface, it managed under few seconds.
Than i request to rewrite the script to add integration with windows explorer to choose the directory and a drop list for the git command to use, it managed under few seconds.
A few errors when i launch the script and it correct under few seconds.
Than i request to rewrite the script to remember the last location and last command , it managed under few seconds
Simple amazing.
I simple start with a request to writing a powershell script that syncronize a local directory on my pc with a git's remote repository , it managed under few seconds.
Than i request to rewrite the script with parameters for local directory and git command to use, it managed under few seconds.
Than i request to rewrite the script to add a graphic interface, it managed under few seconds.
Than i request to rewrite the script to add integration with windows explorer to choose the directory and a drop list for the git command to use, it managed under few seconds.
A few errors when i launch the script and it correct under few seconds.
Than i request to rewrite the script to remember the last location and last command , it managed under few seconds
Simple amazing.
Last edited:
DrFigs
Member
It didn't like when I pointed out the FBI killed MLK, for what it's worth. And then went on to say that the FBI spying on MLK was legal. It conceded that it's not a legal expert when I gave rational counterarguments.
edit: i tried asking it again and it actually gave me a really nuanced answer.
Last edited:
Sakura
Member
Pretty sure you can already do this. Just use whisper or something to turn your voice into text, and pass it to GPT as the prompt.
Omega Supreme Holopsicon
Banned
I tried llama and alpaca locally on my pc, using 7b model but it was way to stupid to be useful. All answers were totally not related to my questions.
I've heard 7B al is dumb, 13B is about chatgpt 3.5, and 30B is slightly smarter. Some are using alpaca.cpp.
Have also heard that the quantization models might run just as fast or faster on a good cpu vs gpu, due to gpus not being optimized for quantized versions.
E-Cat
Member
What a bunch of losers lmao
GnomeChimpsky
Member
Literally old man yelling at The Cloud.
GPT 4 is a lot better at it.
Ozzy Onya A2Z
Member
What did you ask it that caused that mate? Inquiring minds want to know.
AJUMP23
Parody of actual AJUMP23
I work in an environment where they block a lot of stuff. It was good for a week then a week later blocked.
Ozzy Onya A2Z
Member
Just a joke mate. I get corporate firewalls.
By the look of it Copilot isn't enabled for Word 365 in Australia yet. Bummer I wanted to give it a chance to write my R&D grant application for a client.
AJUMP23
Parody of actual AJUMP23
I knew it was a joke. But I never joke. I am always serious.
Lord Panda
The Sea is Always Right
I've been using ChatGPT4 extensively for both personal projects and work. From organising and building Ansible project files, troubleshooting Docker/Docker-compose and a variety of network issues, to writing up high level strategy documents for various projects and activities, ChatGPT4 has been indispensable.
Often I use it to organise my train of thought and to explore whether I'm heading in the right direction, and using it to alleviate some of the grinding and busy work. As someone that works in IT, I better start thinking about another career (a trade?) or branching out into AI development myself.
Often I use it to organise my train of thought and to explore whether I'm heading in the right direction, and using it to alleviate some of the grinding and busy work. As someone that works in IT, I better start thinking about another career (a trade?) or branching out into AI development myself.
Last edited:
CrankyJay™
Member
So small update as a dev. Been using this to help with some Azure pipelines stuff piecing together YAML script etc.
It’s excellent at syntax related stuff, but where GPT seems to struggle is Nuget packages and APIs. The model is dated so unless you specify a version it’s probably assumed it’s a version from 2021, so you’re most likely getting outdated information.
I’ve also seen instances of it outright making up parameters that never existed.
It’s excellent at syntax related stuff, but where GPT seems to struggle is Nuget packages and APIs. The model is dated so unless you specify a version it’s probably assumed it’s a version from 2021, so you’re most likely getting outdated information.
I’ve also seen instances of it outright making up parameters that never existed.
Alx
Member
Except (good) journalists know they have limited knowledge of the fields they're talking about, and try their best to be accurate. While an AI only tries to appear accurate, with no consideration of truth, consequences and deontology.
Grildon Tundy
Member
It is trained on human-created data points, after all. Funny to see an AI getting close to having the sum of all human knowledge, and we see how full of BS it is lol
CrankyJay™
Member
Wonder if this means China and Russia can try to purposely feed it misinfo.
Grildon Tundy
Member
If China and Russia have fed data into what they're using for ChatGPT's data model, and there isn't some magical way for OpenAI to control for that, then it's already happening as a side effect
Last edited:
GeekyDad
Member
Hope this isn't off-topic, but it is AI-related. My turn with Google's Bard came up about a week ago. I'm just now getting around to give it a try. I've still got it opened on another tab. So far, it looks like it's just a chat with the AI. Impressive, though. I entered only a handful of keywords (woman, field, empty, rain, factory, home, gentle, search), and it spat out a really pretty poem.
"The woman stood in the empty field, her face turned up to the sky. The rain was falling softly, and she closed her eyes, letting the drops fall on her face. She took a deep breath, savoring the smell of the wet earth.
She had been working in the field all day, and she was tired. But she felt a sense of peace and contentment. She was home.
She had been born and raised on this farm, and she loved it. She loved the land, the animals, and the people. She loved the hard work, and the satisfaction of a job well done.
But she also loved the quiet moments, like this one. When she could just stand and listen to the rain, and the wind in the trees. When she could just be.
She opened her eyes and looked around. The field was empty, but it was not barren. It was alive with the promise of new life. The seeds that had been planted would soon sprout and grow, and the field would be full of crops.
The woman smiled. She knew that she was part of something bigger than herself. She was part of the cycle of life, and she was grateful for that.
She turned and walked back to her house, her footsteps light on the wet ground. She was home."
"The woman stood in the empty field, her face turned up to the sky. The rain was falling softly, and she closed her eyes, letting the drops fall on her face. She took a deep breath, savoring the smell of the wet earth.
She had been working in the field all day, and she was tired. But she felt a sense of peace and contentment. She was home.
She had been born and raised on this farm, and she loved it. She loved the land, the animals, and the people. She loved the hard work, and the satisfaction of a job well done.
But she also loved the quiet moments, like this one. When she could just stand and listen to the rain, and the wind in the trees. When she could just be.
She opened her eyes and looked around. The field was empty, but it was not barren. It was alive with the promise of new life. The seeds that had been planted would soon sprout and grow, and the field would be full of crops.
The woman smiled. She knew that she was part of something bigger than herself. She was part of the cycle of life, and she was grateful for that.
She turned and walked back to her house, her footsteps light on the wet ground. She was home."
Last edited:
Roni
Member
The AI can learn from any human, Copilot learns from code you've already written and ChatGPT can remember a certain length of conversation before going haywire and acting as if you've never talked about something. They do need to improve the memory aspect of it. I don't mind giving it instructions, but when it fails to hold everything we've discussed "in mind" I find it frustrating.
Here's a conversation I had around the turing test. It's pretty clear towards the end it's programmed not to talk about it. I actually recently felt they dumbed him down a little:
It's pretty great at just having a conversation though, rolls with the punches and tries to make sense of what you actually mean...
Nintendo's Witch
Member
nvm i saw there is even a thread about it lol
Last edited:
01011001
Banned
the guy who's making a really impressive Mario 64 mode, and while doing so basically rewriting almost the entire engine in order to optimise the game, trying to see if ChatGPT 4 can improve his code.
his optimised version of Mario 64 would most likely run at 60fps on original hardware in its current form already. his way levels he made for his mod are way more demanding and still reach 40 to 50fps on og hardware.
so it's really hard to actually improve this further and a really hard problem to ask chat gpt.
he did the same experiment with version 3.5, which was already promising but not quite there yet.
version 4 seems to be way better at coding, but isn't all that amazing at optimising N64 code
but that is to be expected I think lol.
BUT it did spit out 2 decent results!
his optimised version of Mario 64 would most likely run at 60fps on original hardware in its current form already. his way levels he made for his mod are way more demanding and still reach 40 to 50fps on og hardware.
so it's really hard to actually improve this further and a really hard problem to ask chat gpt.
he did the same experiment with version 3.5, which was already promising but not quite there yet.
version 4 seems to be way better at coding, but isn't all that amazing at optimising N64 code
but that is to be expected I think lol.BUT it did spit out 2 decent results!
Last edited:
HoodWinked
Member
Its been a week since this came out but I kept putting off watching it cause watching Lex feels like school work but turned out to be an excellent interview and while I'm still super black pilled on Ai I do have a little hope after watching this. Sam Altman does seem like a very thoughtful and decent dude. Forgot to mention for people that don't know he's the CEO of OpenAi
Last edited:
Ozzy Onya A2Z
Member
I find it interesting over at ERA they have now banned AI topics. Some moderators worse than AI confirmed. Nice to see GAF being open for discussion of topics once again, not just gaming related either.