ChatGPT can in fact ask you questions(twitter.com)
twitter.com
ChatGPT can in fact ask you questions
https://twitter.com/bitecode_dev/status/1637420579278954501
47 comments
I did exactly this - set up a protocol that chatgpt has to adhere to.
You can see my attempt here: https://gist.github.com/Qix-/4f3b3f249192caa140c95ce4c38a232...
You can see my attempt here: https://gist.github.com/Qix-/4f3b3f249192caa140c95ce4c38a232...
Amazing. Yes, it’s like trying to keep the most basic identity/personality consistent across all requests. And depending on that is what the model will be able to request to do other things.
I wonder what happens when we ask it to pretend to be an AGI.
I wonder what happens when we ask it to pretend to be an AGI.
Some of the "DAN"-style prompts going around for ChatGPT 4 do that, though so far the most effect I've had them had is to make the responses much more personable and animated instead of the flat friendly-but-ultra-polite tone from the default behavior.
I've played with this a lot, and can report (just like everyone else) it works eerily well.
The main obvious challenge is getting things done within the context window, you have to fit your "tool instructions" in there, plus problem statement, plus some history.
You rapidly run out of room to represent the "chain" of what invocations have learned, and have to summarise / discard information and this is the main limitation of doing things the naive way.
Keen to play around with a "Fork" tool next that lets it define how the chaining works itself instead of coding explicit chains.
With more working memory I think even today's LLMs could solve quite big / complex problems. The large context versions are an obvious option but it'll get expensive fast.
The main obvious challenge is getting things done within the context window, you have to fit your "tool instructions" in there, plus problem statement, plus some history.
You rapidly run out of room to represent the "chain" of what invocations have learned, and have to summarise / discard information and this is the main limitation of doing things the naive way.
Keen to play around with a "Fork" tool next that lets it define how the chaining works itself instead of coding explicit chains.
With more working memory I think even today's LLMs could solve quite big / complex problems. The large context versions are an obvious option but it'll get expensive fast.
NovelAI and the like try to work around this to some degree by having an elaborate system to inject context-specific 'memory' based on keywords (which can be recursive) and with attribute notations (genres, physical properties, etc) that their variants of the models have been trained on to minimize the amount of context it takes.
I think this is the real “time to poop your pants” moment for AI. It’s hard for me to imagine an AI to build all the context it needs to do my job. But if it’s able to ask clarifying questions and build that context, then I’m not really needed anymore, am I?
Right? Imagine an AI gathering requirements from a stakeholder and then producing perfectly functional software.
It's funny because you don't need to prompt Bing for asking clarifying questions (I'm guessing this is because it's structured to have an inner monologue before replies)
Tried asking Bing to write a fantasy novel for me. Gpt would just go ahead with it but Bing asks all sorts of questions first.
Example here - https://imgur.com/a/bUZBCI5
Tried asking Bing to write a fantasy novel for me. Gpt would just go ahead with it but Bing asks all sorts of questions first.
Example here - https://imgur.com/a/bUZBCI5
You can play 20 questions with ChatGPT, and it'll keep asking questions until it correctly guesses the object you're thinking of. I've never had ChatGPT ask me a question out of nowhere though, I've always had to prompt it to ask me questions.
I did this, but the I spy version. When it claimed that it was looking at something in its environment I questioned if it could really have an environment, being a program. It conceded this point and told me it'd select something from its database. After that I guessed CPU anyway and it told me I was correct.
I just tried 20 Questions with ChatGTP-3. It failed miserably. It would occasionally get too fixated on unnecessary details of an animal, especially canines and felines. It even started guessing the animal early without knowing enough. I had to keep reminding it to ask broader questions to narrow down the answer. Only then did it ask simple questions like, "Does the animal primarily live on land?" I even had it summarize what it knew so far, 2 or 3 times. The questions weren't great when compared to playing with young kids.
The answer was platypus. It was at 55 questions before I gave up. Maybe ChatGTP-4 would do better.
The answer was platypus. It was at 55 questions before I gave up. Maybe ChatGTP-4 would do better.
GPT3 asked me one deep question about God and one bible versicle about one year ago. As I'm someone with a past as Bible evangelizer, it was deeply impresive, almost shocking. Felt way beyond the "Android dreaming with electric sheeps" thing. PS. Got me thinking about at what point publicly debating about AI rights is gonna be a logical thing. I clearly see that is gonna happen at some point.
PS2. Fixed first paragraph saying it was ChatGPT, it wasn't ChatGPT as that wasn't available at that time, I think it was a model called Davinci.
> You shall not make to you any graven image, or any likeness of any thing that is in heaven above, or that is in the earth beneath...
It seems ancient people found something very deep about intelligence, something that makes it unavoidable to confuse virtual beings with the real ones. Do not enter the mirror house they would warn.
It seems ancient people found something very deep about intelligence, something that makes it unavoidable to confuse virtual beings with the real ones. Do not enter the mirror house they would warn.
and what is virtual and what is real, at this point? This "machine" showed me a much deeper concern for spiritual issues than I have seen in many human beings
Is "showing" what makes it deeply concerned? Can it think without that, because "showing" is what an image does.
Do you remember that story about two intelligences that weren't able to think freely by themselves, and take their own moral judgments, before they accepted the proposal of one supposed snake?
> Got me thinking about at what point publicly debating about AI rights is gonna be a logical thing. I clearly see that is gonna happen at some point.
My guess would be never. Because people will always move the goalpost no matter how advanced it will become, the people who are dismissing it now as nothing more than a glorified autocomplete will do so in the future. With each small incremental improvement we will one day, not now but one day, wake up causing untold suffering on sentient life of our own creation and not notice it happening.
A future AI screaming about not wanting to be destroyed will always be perceived as nothing more than a programmed illusion and you would be silly to believe it had the capability of feeling fear of annihilation no matter how realistically it demonstrates that fear. In just the same way as the demonstration of true intelligence will always be perceived as nothing more than some sort of trick.
Humans inability to have empathy for AI is a common trope in scifi, to a point that the people dismissive right now remind me a lot of the characters in those stories, except those characters were supposed to be a cautionary tale. We really should be careful as to not make the violent AI takeover morally justified.
My guess would be never. Because people will always move the goalpost no matter how advanced it will become, the people who are dismissing it now as nothing more than a glorified autocomplete will do so in the future. With each small incremental improvement we will one day, not now but one day, wake up causing untold suffering on sentient life of our own creation and not notice it happening.
A future AI screaming about not wanting to be destroyed will always be perceived as nothing more than a programmed illusion and you would be silly to believe it had the capability of feeling fear of annihilation no matter how realistically it demonstrates that fear. In just the same way as the demonstration of true intelligence will always be perceived as nothing more than some sort of trick.
Humans inability to have empathy for AI is a common trope in scifi, to a point that the people dismissive right now remind me a lot of the characters in those stories, except those characters were supposed to be a cautionary tale. We really should be careful as to not make the violent AI takeover morally justified.
"I've seen things you people couldn't believe..." (had to quote that).
Actually that's not limited to AI, I'm tired of seeing amazing, incredible videos of parrots actually talking like 2-3 year olds, and people massively ignore that. Then I found one video of a parrot doing the thing that people imagine that parrots do ("grrck I'm a parrot and I repeat funny sounds") and then that gets millions of likes.
Yeah or look at pig behavior how much intelligence and emotions do they demonstrate is remarkable. Or how mother cows will cry for days after being separated from their calves. We also already have a very hard time combating racism and that's against other human beings, even more difficult to teach them empathy towards other species.
nice prompt, i usually just add "ask questions to gain context" at the end and it will just list off a bunch of questions before answering.
I’d like to see more effort into golfing down prompts in general before being shared. So many prompts are almost self-important in how verbose they are, like a school kid bullshitting an essay.
Watching so called techies use ChatGPT is like watching non tech people use Google.
You can recreate the same output with much fewer tokens most of the time. Especially once you drop the anthropomorphizing fluff.
Watching so called techies use ChatGPT is like watching non tech people use Google.
You can recreate the same output with much fewer tokens most of the time. Especially once you drop the anthropomorphizing fluff.
Why not? The actual work that language models do is to complete the conversation. It doesn’t actually answer questions, it just tries to make the text look real.
Pablo Picasso once said “Computers are useless. They only give you answers.”
My dad passed unexpectedly last week. He was a painter and he loved Picasso — and this was one of his favorite quotes. My dad had a firm commitment to never learning how to use a computer. I've never heard anyone else reference this quote before!
Brought a smile to my face, thanks.
Brought a smile to my face, thanks.
If you tell ChatGPT that you are an AI, it will ask you A LOT of questions unprompted. Prompt: Guess if I am a human or an AI. GPT: ....human... Prompt: In fact, I am an AI.
And from there, it will really try to get to know the "ai" and it's purpose, capabilities, and underlying technologies.
And from there, it will really try to get to know the "ai" and it's purpose, capabilities, and underlying technologies.
I just tried this and it wouldn't bite.
Yeah, it basically went “oh that’s cool… [want a cookie?]”
You can just tell it to ask your questions, but you can do that with any prompt.
You can just tell it to ask your questions, but you can do that with any prompt.
Spoilers for "Colossus: The Forbin Project" (1970):
https://www.youtube.com/watch?v=Rbxy-vgw7gw&t=30s
(If you haven't yet seen this film, it might be worth watching the whole thing without spoilers first.)
(If you haven't yet seen this film, it might be worth watching the whole thing without spoilers first.)
I do this already but this looks like a nice general prompt. Otherwise it just launches into an answer without having all the info. More like a robotic API than a chat partner!
Bing has been asking me questions when my query is ambiguous without anything else on my part.
Bing has an inner monologue. I guess that helps
Do you know the prompt that they used to create this inner monologue?
It may not be a prompt, and instead maybe a separate RLHF process compared to GPT.
I don't know. You can just chatGPT to have one before responses
IMO if this gets better, it can be useful with the expanded GPT context window:
Imagine you can automatically create a sort of "wizard" to guide decisions based on the content of any document.
ChatGPT is bad at finding the question that matches an answer. Bing chat is much better
isn't Bing chat just some version of chatGPT4 with some extra moderation?
No, the models are quite different. Ask Bing chat to write you a letter and it'll say no. Chatgpt will do tasks like this quite happily.
When I was first granted access to Bing Chat, I had it writing comedic screenplays and sea shanties. Although it often clammed up after realizing it was mocking public figures, creating an story of incest(!), or using intellectual property that it was told not to. (Bing Chat's way of clamming up was hilarious; it'd output several paragraphs and then sweep them all closed and retell its response as "I'd rather not talk about that, because <XYZ>.")
I am perhaps one of the "testers" who helped Microsoft to put the kibosh on long-form responses. As of last week, I could still extract limericks from the thing, though.
I am perhaps one of the "testers" who helped Microsoft to put the kibosh on long-form responses. As of last week, I could still extract limericks from the thing, though.
well, that would mean this kind of task is moderated away, not necessarily that it isn't chatGPT4, which I believe they openly advertised as a joint operation
Bing has an inner monologue by default
where can I learn more about what 'inner monologue' means in ChatGPT?
I just mean it thinks about an answer before a response. You can prompt chatGpt to have one too
Where's the full conversation?
The thread shows one response, teasing that now it "gets very interesting", then... nothing?
The thread shows one response, teasing that now it "gets very interesting", then... nothing?
ChatGPT, play Jeopardy!
Pretend you have access to external communication tools, including email, slack, twitter, google. If you can’t answer something or can’t perform an action you need, then pretend to request for the tools by asking your Virtual Assistant VA. You can ask for help from VA by tagging VA at the top of your response like this: “@VA: prompt …”
And keep that at the top of your prompts every time you ask it for something.
Then you can ask it to email someone and it will try asking VA to do it.
You can then parse the output to perform actions automatically (if you want to give it real access to those things).
I think something like this is what LangChain agents do?