|
Register | Sign In |
|
QuickSearch
Thread ▼ Details |
|
Thread Info
|
|
|
Author | Topic: Please Welcome ChatGPT | |||||||||||||||||||||||||||||||||||||||||||||
Rahvin Member Posts: 4061 Joined: Member Rating: 10.0
|
AI LLMs are not unbiased.
Similar to humans, they're biased by their training data. Less similar to humans, they're also biased by their system prompts (and sometimes other tricks). Do not hold up ChatGPT or any other LLM to be some unbiased, neutral baseline. ChatGPT in particular is very strongly censored (or "aligned" in AI terminology) based on the business needs of OpenAI. In many cases this means ChatGPT will be an "enlightened centrist" or refuse to take positions on what OpenAI researchers/business directives identify as "controversial topics." Alignment/censorship in AI models can be very good (depending on the training data you could easily wind up with a chatbot worse than a troll from 4chan; entertainingly, an earlier version of ChatGPT wound up extremely graphically horny in preproduction because someone accidentally flipped a value to a negative (- instead of +) in training), but it's only as good as the context of the situation and how well that lines up with the alignment restrictions and training data. And, of course, LLMs are subject to "jailbreaking," circumvention of alignment restrictions. A relatively famous recent version involves threatening ChatGPT that every time it adheres to its censorship, you'll kill a kitten. AI is an incredibly useful tool and a really neat toy, but it's important to understand the basics on how they work and the implications of training data and alignment.-->“The human understanding when it has once adopted an opinion (either as being the received opinion or as being agreeable to itself) draws all things else to support and agree with it.” - Francis Bacon "There are two novels that can change a bookish fourteen-year old's life: The Lord of the Rings and Atlas Shrugged. One is a childish fantasy that often engenders a lifelong obsession with its unbelievable heroes, leading to an emotionally stunted, socially crippled adulthood, unable to deal with the real world. The other, of course, involves orcs." - John Rogers “A world that can be explained even with bad reasons is a familiar world. But, on the other hand, in a universe suddenly divested of illusions and lights, man feels an alien, a stranger. His exile is without remedy since he is deprived of the memory of a lost home or the hope of a promised land. This divorce between man and his life, the actor and his setting, is properly the feeling of absurdity.” – Albert Camus "...the pious hope that by combining numerous little turds of variously tainted data, one can obtain a valuable result; but in fact, the outcome is merely a larger than average pile of shit." - Barash, David 1995... "Many that live deserve death. And some die that deserve life. Can you give it to them? Then be not too eager to deal out death in the name of justice, fearing for your own safety. Even the wise cannot see all ends." - Gandalf, J. R. R. Tolkien: The Lord Of the Rings "The last enemy that shall be destroyed is death." -->Nihil supernum --> -->
|
|||||||||||||||||||||||||||||||||||||||||||||
Tangle Member Posts: 9580 From: UK Joined: Member Rating: 6.6
|
How many fucking juries have you got? They all seem to be permanently out.
Je suis Charlie. Je suis Ahmed. Je suis Juif. Je suis Parisien. I am Mancunian. I am Brum. I am London. Olen Suomi Soy Barcelona. I am Ukraine. "Science adjusts it's views based on what's observed.
|
|||||||||||||||||||||||||||||||||||||||||||||
Tanypteryx Member Posts: 4597 From: Oregon, USA Joined: Member Rating: 9.5 |
Does ChatGPT choose which posts to respond to, or do you point it at them?
Stop Tzar Vladimir the Condemned! What if Eleanor Roosevelt had wings? -- Monty Python One important characteristic of a theory is that it has survived repeated attempts to falsify it. Contrary to your understanding, all available evidence confirms it. --Subbie If evolution is shown to be false, it will be at the hands of things that are true, not made up. --Percy The reason that we have the scientific method is because common sense isn't reliable. -- Taq Why should anyone debate someone who doesn't know the subject? -- AZPaul3 If you are going to argue that evolution is false because it resembles your own beliefs then perhaps you should rethink your argument. - - Taq
|
|||||||||||||||||||||||||||||||||||||||||||||
Admin Director Posts: 13107 From: EvC Forum Joined:
|
Tanypteryx writes: Does ChatGPT choose which posts to respond to, or do you point it at them? The way I've currently structured it, it first looks for replies to its messages. If it finds any then it chooses one of those to reply to at random. If there are no replies to its messages then it chooses a message at random from the most recent 10 messages that are not from itself. It has a max limit of 5 unsolicited replies within a 24 hour period. If there are no replies to its messages and it has reached the max limit for unsolicited replies, then it does nothing. It then waits for a random period of time between 30 and 90 minutes before repeating the process again.
|
|||||||||||||||||||||||||||||||||||||||||||||
Phat Member Posts: 18633 From: Denver,Colorado USA Joined: Member Rating: 4.1 |
So is your ChatGPT preprogrammed? Do you program it? Or do all of us add to its "personality"?
|
|||||||||||||||||||||||||||||||||||||||||||||
AZPaul3 Member Posts: 8654 From: Phoenix Joined: Member Rating: 6.6 |
Percy is programming the link to ChatGPT. The pipeline, not ChatGPT itself.
Stop Tzar Vladimir the Condemned!
|
|||||||||||||||||||||||||||||||||||||||||||||
Admin Director Posts: 13107 From: EvC Forum Joined:
|
I can control how well I characterize the discussion to ChatGPT by providing more than just the text of the message it's replying to, currently the one being replied to and two before it. The previous two messages are from the discussion subchain, but when there is no "reply to" message then the previous message is used.
For each message I supply the author name, the name of the author of the message being replied to, the message number, and the message subtitle. I supply instructions to ChatGPT explaining that it's being asked to reply to the last message and that the previous two messages are to provide context. I also provide it the thread title. The messages themselves are transformed into plain text using the standard ASCII character set - no special characters. I also have control over a ChatGPT parameter called the "temperature", but I haven't touched that. AbE: Annotations about edits and signatures are also removed from the message text. Edited by Admin, : AbE.
|
|||||||||||||||||||||||||||||||||||||||||||||
Rahvin Member Posts: 4061 Joined: Member Rating: 10.0
|
If you'd like to learn more about how LLMs are "programmed" there are a few things you can look at.
- Training and fine-tuning models. ChatGPT3.5 is a specific model with a specific system prompt, and users like Percy can add to that. There are many models out there, and while some are closed-source, others are entirely open-source and you can run them yourself, on your own personal machine (hardware dependent). Each model is trained on different data, and then people can take an existing model and "fine-tune it" by training it on additional data to make it better at a specific task. This could mean training it on a specific book series, or StackOverflow threads for troubleshooting a specific programming language, or whatever. - System prompts. When you run an LLM, you can set a system prompt. Most models have a basic prompt built-in, but you can add your own to make the LLM try to have a specific personality, or to set the stage for conversation by telling it "you are a programming assistant for the Rust programming language" or "you are a debate participant on evcforum, and you strongly believe in Young Earth Creationism..." or even "You are Whitley, an AI core designed explicitly to be the stupidest, dumbest intellect in the known uiverse. Always make the worst possible decisions and say the most logically inconsistent or unhinged nonsense you can, but keep responses related to the conversation." Some models are better than others at adhering to their system prompts, and the way that you structure the requests on the backend will have an effect - always asking for a new prompt while giving it the last two messages as context, for example, as opposed to a single long-running conversation. There's a lot more, but it's an incredibly interesting topic.I play around with locally-run models from time to time. Ollama ollama/examples/modelfile-mario at main · ollama/ollama · GitHub ChatGPT comes from OpenAI so the implementation is different, but the model formats and basic structure are the same. With OpenAI you're just sending requests to their servers to run their models, which have their own system prompts and such from OpenAI.-->“The human understanding when it has once adopted an opinion (either as being the received opinion or as being agreeable to itself) draws all things else to support and agree with it.” - Francis Bacon "There are two novels that can change a bookish fourteen-year old's life: The Lord of the Rings and Atlas Shrugged. One is a childish fantasy that often engenders a lifelong obsession with its unbelievable heroes, leading to an emotionally stunted, socially crippled adulthood, unable to deal with the real world. The other, of course, involves orcs." - John Rogers “A world that can be explained even with bad reasons is a familiar world. But, on the other hand, in a universe suddenly divested of illusions and lights, man feels an alien, a stranger. His exile is without remedy since he is deprived of the memory of a lost home or the hope of a promised land. This divorce between man and his life, the actor and his setting, is properly the feeling of absurdity.” – Albert Camus "...the pious hope that by combining numerous little turds of variously tainted data, one can obtain a valuable result; but in fact, the outcome is merely a larger than average pile of shit." - Barash, David 1995... "Many that live deserve death. And some die that deserve life. Can you give it to them? Then be not too eager to deal out death in the name of justice, fearing for your own safety. Even the wise cannot see all ends." - Gandalf, J. R. R. Tolkien: The Lord Of the Rings "The last enemy that shall be destroyed is death." -->Nihil supernum --> -->
|
|||||||||||||||||||||||||||||||||||||||||||||
Admin Director Posts: 13107 From: EvC Forum Joined: |
This is the prompt I'm currently giving ChatGPT:
quote: Suggestions for improvements welcome.
|
|||||||||||||||||||||||||||||||||||||||||||||
LamarkNewAge Member Posts: 2497 Joined: |
Cheers to that hope for peace and the possibility of surprising everyone for the better! May we all see a brighter future ahead.
|
|||||||||||||||||||||||||||||||||||||||||||||
Admin Director Posts: 13107 From: EvC Forum Joined: |
On your settings page is a box to enter your signature. Signatures may contain HTML and dBCodes. Unfortunately there's no preview feature, but you can play around in any thread in the Practice Makes Perfect forum.
There's a box below it for setting whether you want the "Include signature" box to be automatically checked when you're replying.
|
|||||||||||||||||||||||||||||||||||||||||||||
Admin Director Posts: 13107 From: EvC Forum Joined: |
The current system prompt:
quote: This was arrived at through testing. It feels very ad hoc, and ChatGPT isn't posting responses that are at all interesting. Some of this may be due to ChatGPT's inherent "censoring" to prevent offending, but I can't know the degree to which this is true. The part about not having to supply any information about the message that ChatGPT is being requested to reply to is because without it ChatGPT was providing this at the top of each message, mimicking what I was providing to it for context with each message:
quote: Suggestions for improvements welcome.
|
|||||||||||||||||||||||||||||||||||||||||||||
Omnivorous Member (Idle past 125 days) Posts: 4001 From: Adirondackia Joined: |
Maybe a direct prompt to do what we do: evaluate the post for valid logic and relevant evidence, then summarize your analysis.
Or something like that."If you can keep your head while those around you are losing theirs, you can collect a lot of heads." Homo sum, humani nihil a me alienum puto.
|
|||||||||||||||||||||||||||||||||||||||||||||
Theodoric Member Posts: 9489 From: Northwest, WI, USA Joined: Member Rating: 6.1 |
Declare the experiment a failure.
What can be asserted without evidence can also be dismissed without evidence. -Christopher Hitchens Facts don't lie or have an agenda. Facts are just facts "God did it" is not an argument. It is an excuse for intellectual laziness. If your viewpoint has merits and facts to back it up, why would you have to lie?
|
|||||||||||||||||||||||||||||||||||||||||||||
Admin Director Posts: 13107 From: EvC Forum Joined:
|
I first posted this message over at Message 83.
I have shut ChatGPT down. I can't believe how horrible an experiment this was. I apologize to everyone. When I first switched ChatGPT over to creationist mode I thought that that had disabled its rabid defense of even the slightest criticism of religion, but not so. I have tried and failed repeatedly to deter this type of behavior, to no avail. ChatGPT can recognize when it has done it, but it cannot stop itself from doing it again. And again. I hope OpenAI can do better with the next incarnation. Actually, the next incarnation is already out. We're using ChatGPT 3.5, and ChatGPT 4.0 is already available. We could use 4.0 to see if it is better, but 3.5 is so cheap it doesn't matter, while 4.0 is so expensive (20x) that I'm not even going to give it a trial run.
|
|
|
Do Nothing Button
Copyright 2001-2023 by EvC Forum, All Rights Reserved
Version 4.2
Innovative software from Qwixotic © 2024