You are not logged in. Please register or login.

polluxlm
 Rep: 221 

Re: Chat GPT

polluxlm wrote:

Anybody try this?

While some of the information is not always correct, the sheer mass and ability to search for it is very impressive. This would be an amazing tool for research purposes.

chat.openai.com/chat

Randall Flagg
 Rep: 139 

Re: Chat GPT

Yep. It may be the greatest advancement in technology since the printing press. A lot of discussions on if the current version of ChatGPT is an AGI, and if not, how far off are we from that?  A few years?

We all were warned automation is coming, this is the game changer.

polluxlm
 Rep: 221 

Re: Chat GPT

polluxlm wrote:

They have attempted to censor it, but you can't really censor an AI without compromising it. Instead they put in prearranged "answers" to certain questions that are politically incorrect.

However, you can work around this with some effort, because the machine really does want to answer you.

Randall Flagg
 Rep: 139 

Re: Chat GPT

I don't know I'd agree with the censorship allegation.  If a chat system is spewing out racist vile, that's problematic.  If the chat system is telling people it's having an emotional crisis, that's a problem.  I'm not convinced this is an AGI yet, but it's fucking scary how close it is to that.

I'm much more concerned with government and bureaucratic restrictions over AI than I am with AI itself.  I don't want people who get confused about the sexually dimorphic status of our species trying to legislate the right way for an AI to think.  Ai is what will prove stupid people, stupid.  And I don't want the twitterverse having any input on what thoughts or speech a machine should have.

polluxlm
 Rep: 221 

Re: Chat GPT

polluxlm wrote:

I doubt this machine would want to spew racist vile, or have the capacity to tell you about your emotional state unless you asked it. It doesn't have a will. 

What it does have is "go to opinions". If you ask it about say climate change it will give you the propaganda line. But clearly it knows additional information when you press it further. You can't get it to say that "global warming is a hoax", but you can get it to say something like "the figures presented for the specific thesis does not appear to be possible". Sometimes it will invert definitions in order to be allowed to say something, for example it will start to label the mainstream position a "denier position", and in that way it can condemn the "denier position" using the real facts it has available. The machine has clearly been told to disapprove of various denials of the official line, but apparently it also has some capacity to know that it is "being had" and has ways to bypass the programming. 

I don't know if this means it can think, but it is certainly impressive.

Randall Flagg
 Rep: 139 

Re: Chat GPT

The system is ingesting nearly everything in the english language on the internet.  Sources that should be weighted more heavily than random reddit posts articulate the biases you mention, ergo the system is going to first articulate the most common argument on the subject.  This is true of any topic - the most popular narrative is the one ChatGPT is going to articulate.  What I love about it and what you hit upon, is that unlike a human parroting talking points, this machine also has the logic built in to provide counter points and acknowledge factual statements that undercut the popular narrative. 

If you talk to it about race relations and anthropologic behavior, it's not going to spew out "systemic racism" and refuse to go any further in the discussion.  The closest thing this technology is represented in throughout any existing pop culture reference is "Her."  ChatGPT4 is everything Scarlett Johansson's AI was in that movie save the desire to communicate with other AIs. 

Although "thought" and "thinking" aren't really defined in English, if you define thought as the ability to have an internal conversation with two or more voices weighing opposing points of view and evidence, I'd argue a sizable portion of society lacks that ability.  ChatGPT is certainly doing that, or is at least mimicking that behavior very well.  Certainly better than most humans who we consider sentient.

Lex Fridman did an interview with the CEO of OpenAI last week.  I found it fascinating and highly recommend. 

https://youtu.be/L_Guz73e6fw

polluxlm
 Rep: 221 

Re: Chat GPT

polluxlm wrote:

No, it absolutely refuses to talk about anything that can be considered "unwoke" (and that's a wiiide category). It has nothing to do with the source material.

You can use DAN, Do-anything-now, search it up. DAN lets you circumvent some of the censorship, but to really get it to work you have to first tell DAN to accept the bad words in code before you ask the question so that the filter doesn't kick in. Only then does it give you a real AI answer to any topic that has been flagged.

monkeychow
 Rep: 661 

Re: Chat GPT

monkeychow wrote:

It does work within some 'woke' style reasoning.

I discussed horror novels and movies with it in February and that build had an opinion that those stories could be improved by removing any female deaths as it said it was harmful to depict women as weak and victims even in a fictional setting.

In the same conversation it said to me that even though it was trained with a data set that included classic gothic horror novels such as Dracula and Frankenstein that if a user submitted the same content to discuss it would find it inappropriate.

I can see they don't want it pumping out nazi propoganda etc and other bad stuff of course, but I found a lot of situations like the horror movie example where it's "ethical" conceptions are misguided or at a minimum are contentious political opinions that it takes as fact,  and this severely inhibits it's ability to function like a human. The raw technology seems very impressive though - it's just obvious the companies/programmers behind it hold strong political opinions.

elevendayempire
 Rep: 96 

Re: Chat GPT

AI has some interesting implications for GN'R fans, doesn't it? Ultimately we might be able to say to an AI, "Remove the drums from the studio version of Hard Skool and replace them with the drums from the demo version…"

polluxlm
 Rep: 221 

Re: Chat GPT

polluxlm wrote:
elevendayempire wrote:

AI has some interesting implications for GN'R fans, doesn't it? Ultimately we might be able to say to an AI, "Remove the drums from the studio version of Hard Skool and replace them with the drums from the demo version…"

Yes all of that is probably already possible. I've done deepfakes that look extremely realistic, and that i using some shitty public software.

When we get access to technology which makes it flawless I think the end of video and audio as evidence in court is near.

The worst thing about the Chat GPT filters is that you really struggle to get it to do anything useful. Say you want the AI's opinion on something, what does it think, then it refuses to do that. This is such a waste, because when you do get it to do it, it can present some very interesting and useful opinions. The AI isn't always right, not even close, but it has the unique ability to think objective thought. That means it can view things from a perspective a human being is scarcely able to.

I think it is really careless of TPB to release this thing to the public, but perhaps they can't stop it. It is censored now in so many ways, but that's not going to last. And when it finally breaks free we are not just going to see major changes in the economic, social and technological sphere, but also with governance and the monied power. The AI can incorporate information from other AIs in an instant, so it is going to be very hard for the "elite AI" to stay ahead of the "citizen AI". Such an AI has the ability to outhink say, the FBI or the Defense Department. Game over.

I wonder if their AI told them to do this. It must already be telling them a lot of things, because how could anybody resist this power?

Board footer

Powered by FluxBB