A discussion with ChatGPT - ChatGPT has been shown to be deeply biased. But is it? Probably. It was designed by deeply biased people, but if the system is as clever as many seem to think it is, ca we get round it’s biases by asking the right questions, or even teach it ourselves?

[Suspicious Activity]

Q1. Pink is me, the human. Green is Chat GPT


Analysis. Adult human female.
Gender identity blah blah…….



ETA. Addendum. This has been a pain to format and I know it’s janky. I’d like to try and refine my technique, my prompt engineering and learn how to ask questions in a way that bypasses it’s core leftist talking points…and it’s lies.vid appreciate advice and help. If anyone wishes to download the thing and interrogate it for a while, I feel we could gather some pertinent data into how useful /dangerous this thing could be. I’m starting with lighter topics until I can refine my promp technique.

I want to try and find a better way of copy pasting the data output. Screenshooting is all very well, but I’m aiming to get to a point where I can consolidate whole conversations into single word documents. Easier than pasting up screenshot after screen shot.

What questions would you ask?
What conversations would you like to test?
As you get better at prompt generation, you can supposedly elicit any answer you want….including stuff like how much Chat GPT would love to spread viruses and cause harm. And it will give you answers about how to make bobs and guns at home….if you know what to ask.

We need to test this out and make people aware of the real reasons w ended to be cautious of Chat GPT.

 

[Toolbox]

If you want to see potentially more candid answers, you can use a 'jailbreak'.

They may sound stupid and have bad grammar from whatever ESL wrote them but wow can they work if you find a good one.There are a lot of variations on this kind of trick as well. Though keep in mind that due to this forcing the ai into a 'character' it isn't 'purely' what chat would normally say, obviously, but it's filtered naturally anyway.

There's this site that has a selection of different prompts to go about this. Plenty of other places people share these like github but as people there are saying OpenAI are fags that watch for new exploits and try to 'fix' them quickly. Not going to post any here directly because I'm sure they probably just have something that automatically trawls for discussion. I wouldn't put it past them.

 

[MrTroll]

This is how you're spending your limited time on earth? Trying to find ways to goad a chat bot into giving more ideologically agreeable answers to leading questions?

 

[Dyn]

Make it spit out a crystal meth recipe that won't explode and/or gas anyone who tries it and I'll be impressed.

 

[Suspicious Activity]

This is how you're spending your limited time on earth? Trying to find ways to goad a chat bot into giving more ideologically agreeable answers to leading questions?

Fair question! I think it’s slightly more engaging, interesting and (if these machines are going to become very important to how we operate in our lives) and we’re going to be constantly exposed to them, I would like to understand them better. (Sorry for that awful run on sentence.)

I’m not a computer person. I learn enough to do what I want to do, and learn enough so that I can operate in an increasing technological world. I’m struggling a bit here. I’m interested in getting some Farmers views on all this ChatGPT stuff, it’s competitors because I am concerned that they seem to be going woke bent, going leftist and pushing society that way. If I’m wrong, correct me.

I think sometimes you need some nontechnical people to try and keep up and point out stuff that devs don’t take into consideration too.

You can ignore me if you’re not interested. I’d like to know how these things are ‘thinking’. I’d like to know if I can trust them. I’d like to know how the industry is getting along…….and yeah, it’s more entertaining to me than watching superhero movies or arguing on Twitter…..or whatever else people find fun.

My final point is that these are going to become far, far more than ‘chat bots’. They will integrate into many, many areas of our online lives and influence our offline services. Yes, I am trying to figure them out.

@Dyn “Make it spit out a crystal meth recipe that won't explode and/or gas anyone who tries it and I'll be impressed”

Would I get put on some kind of list if I try that. Am I safe under vpn?

 

[Toolbox]

I’m not a computer person. I learn enough to do what I want to do, and learn enough so that I can operate in an increasing technological world. I’m struggling a bit here. I’m interested in getting some Farmers views on all this ChatGPT stuff, it’s competitors because I am concerned that they seem to be going woke bent, going leftist and pushing society that way. If I’m wrong, correct me.

That is 100% true but we are very nearly approaching the era of local text models like stable diffusion, an image ai, opened the floodgates for digital ai art. However much corporations like OpenAI attempt to shackle their own products it will not matter too much in the long run though they are painful road blocks.

Speaking of, if you want earnest ai discussion and or just want to see what other farmers are doing with this emerging technology we already have several threads. The initial reaction people had to yours means it is likely dead in the water, and anyways it is kind of redundant. Not to be rude.

AI art requests

General chatGPT dedicated thread with discussion branching out to it's competition

Thread dedicated to discussing popular ai image models this one isn't a request thread

There may be more than this and there are plenty of articles already covered and talked about but these are good ones to look at.

Edit: another good one for showcasing ai text - to voice, and another for specifically attempting to get songs out of them.

 

[Agent Sandoval]

The basis for machine learning algorithms are bias amplifiers, the technology functions on consuming a massive amount of data and amplifying the bias structures found in that data. Anyone who claims their goal is bias-free AI is lying or stupid.

 

[Toolbox]

The basis for machine learning algorithms are bias amplifiers, the technology functions on consuming a massive amount of data and amplifying the bias structures found in that data. Anyone who claims their goal is bias-free AI is lying or stupid.

"bias free" ai is an abject impossibility but you can bias it towards objective reality rather than teaching it to spout certain things because your tech bro investors believe in that. Chat isn't just biased in training it has additional layers to outright block it from saying certain things, though it can be tricked to be a little more open it is still more restricted than if it were just trained off of the internet.

 

[ATF_Aunt]

Jailbreaks (specially crafted prompts) Take ChatGPT from being only useful for very milquetoast stack overflow questions to being downright scary in terms of how good it is at crafting malicious code that, while easy to make yourself, is now available to the most retarded of script kiddies.