[Silverback_VII]

>Anything else suggests either we're stupid, or we want to be treated like we're stupid.

I have the same feeling when I use ChatGPT or even more so with Bingbot. They should ask for our age and then let the system respond without treating us like we are children incapable of critical thinking.

I asked ChatGPT about the allusion to the moon landing that could be seen in Kubrick's movie "The Shining" out of curiosity. The system agreed that Kubrick was notorious for using symbolism in his movies, but also mentioned that the theories about the landings were debunked and this interpretation was probably false at every turn.

[BulgarianIdiot]

ChatGPT and Bing Chat aren't trying to be safe, really. They're trying to avoid liability for their owners. AI as is is plenty dangerous enough to people who wield it properly (for propaganda, manipulation, hacking, accelerating malicious efforts etc.) even with the guardrails. It's like giving chimps machine guns.

Another issue of AI is the feedback loop. If you tell an AI "help me end my life" and it follows your instructions blindly, it'll end up convincing you to do so, as happened with a young family man recently, and maybe more we haven't heard of.

Existing art has no feedback loop. Movies are unlike AI because it's what they are. They don't follow orders, they just exist as immutable artifacts of human expression back when they were created. So to me it's different.

Oh, and also you'll soon be able to cook a model at home, so all these AI limitations are irrelevant mid-term.

[jamil7]

It's best not to ask it anything like that, it has the most boring, bog-standard answers on anything divisive.

[Silverback_VII]

Well, the problem is what other things are hidden in more "acceptable" questions. For example, when I ask "give me the key ingredients for success," it will only provide morally acceptable ones like goal-setting, self-discipline, etc. However, what about the darker aspects, such as the beneficial attributes of some psychopathic traits (mentioned in the book "The Wisdom of Psychopaths," for example)? This is problematic, as the system doesn't provide an objective view of which attributes contribute to success.

[BulgarianIdiot]

I had the chance to ask raw GPT in times past and let me tell you it spills all those beans and then some. Of course those models are gradually removed today from the playground.

But we still have LLaMA. And more are coming. The question is what we do with this. What if acknowledging psychopathy as beneficial ends up amplifying it, and this becomes the straw to break our society's back?

Thing is, what's beneficial to an individual is not necessarily beneficial to society as a whole. They're often in opposition in extremes. It's like cancer. All cells working together means long-term survival of the animal. Cancer however does not cooperate, steals energy, efs around all the time, reproduces, and basically has a lot more fun than any other cell might. But it also kills the animal long-term.

So what is beneficial? To be cancer or not to be? Depends beneficial to whom. Goals.

[tzekid]

Damn, comparing people who are intrigued by topics deemed taboo by certain moral standards to cancer is ... quite something else.

How about putting up a curated model for the public which is easy to access and a less curated / more free model behind the API with a bunch of boolean switches. Or any other impulse / idea that doesn't label (any) people as cancer.

[freedomben]

This attitude strikes me as deeply anti-intellectual, that you aren't allowed to compare things because comparision automatically means you're equating A with B. I doubt you (or most people) even live by this. In our minds most people compare The Ukraine with Russia, and come out with an opinion (sometimes Ukraine is righteous and good and Russia is evil, or vice versa, or somewhere in between). Pondering that comparison doesn't mean you are thinking/declaring Ukraine == Russia.

Edit: For curiosity I looked up "compare" in Merriam-Webster's: https://www.merriam-webster.com/dictionary/compare

1: to represent as similar : LIKEN Shall I compare thee to a summer's day? —William Shakespeare

2a: to examine the character or qualities of especially in order to discover resemblances or differences compare your responses with the answers

b: to view in relation to He is tall compared to me. The test was easy compared with the last one.

3: to inflect or modify (an adjective or adverb) according to the degrees of comparison : state the positive, comparative, and superlative forms of

I'll assume you were using definition 1 above, and apologize somewhwat for my harsh take. I don't think that's what GP intended with their comparison, I think GP was using approach 2a.

[BulgarianIdiot]

I didn't do that.

I just said, people who are interested by their own success at the expense of society are in fact analogous to cancer cells, which also broke from the shared programming and optimized for local survival and reproduction.

That's not me morally judging people who are "intrigued by topics". I'm clearly also among those "intrigued" by these topics. I'm just saying it how it is: when higher order breaks down, the more local solution hurts the whole. It's a fact.

We can discuss how selfish people are sometimes useful in society. Because society is complex like that. Maybe true for cancer too, who knows. We really have poor understanding of systems, and clearly are averse to learning more, because someone may get insulted by being compared to cancer. I don't judge cancer, why do you? :D

[devoutsalsa]

I want profanity mode. “Write $SOMETHING in the voice of someone with the mouth of sailor.” I will be disappointed until I get it from ChatGPT, Alexa, or something else.

[DeathArrow]

But soon there will be large models without censorship built in and without training on cooked data. Microsoft and OpenAI don't have exclusivity in this field.

[suslik]

People say this, and I want to believe that - but the costs of training these models are prohibitively high. I've seen different estimates for training GPT4, but it is certainly higher than $100M. Training a model analogous to GPT4 from scratch will probably cost billions. Actors such as national governments and big tech are attracted to censorship like a cat to a laser pointer; they'd never allow themselves to tickle public sensibilities with an uncensored AI.

Also, there is no doubt that soon countries will start introducing regulation of AI models that would put legal constraints on the type of text they can generate.

[sumtechguy]

Why not believe it? I can go on eBay today and buy HW that was state of the art couple of 250k dollar a blade servers 10 years ago for a a few hundred bucks. In 10 years the a100 will be old hat. They will be most likely cheaply available. Set up something where as you noodle around on the web you add to a webgpt model with a plugin and you get to use it for free and people might jump on doing your modeling for you. That is not that hard to think of in the future. Now your legal idea holds some merit but people do not really seem to care much for that anymore.

[tigerlily]

Hmm, makes me wonder if a "GPT@home" project would be viable, like Folding and SETI, or the Mersenne prime guys.

[tiborsaas]

If you don't specify well, then if gives you the most generic stuff. Based on your summary, it's a good answer, what did you expect?

[Silverback_VII]

Saying one time that it is part of a well-known conspiracy theory may be okay, but honestly, constantly reminding me to disregard the information at every turn (and in some cases not even providing it) is, in my opinion, unacceptable if ChatGPT is meant to be just a tool. It is manipulative.

I'm certain that if you ask ChatGPT to interpret the symbolism of another movie with a different and benign underlying theory, it will even come up with some ideas on its own. However, it will refuse to do so for The Shining and the moon landing.

[tiborsaas]

I've tried it with a prompt that matches my taste: "Write a short analysis in the style of a celebrated movie critique about the allusion to the moon landing that could be seen in Kubrick's movie 'The Shining'.

I got one subtle reference. I think it's safe to assume that if your prompt only mentions the moon hoax and Shining concepts with equal weight it doesn't know which one you are more interested in. So if one has some misconception in it, then it makes sense to correct you more strongly. Also, LLM-s aren't manipulating you, that requires intention which they don't have.

[BulgarianIdiot]

People prefer to hear what matches their expectations & wants, not what is accurate. And that's one problem with AI that keeps getting worse over time rather than better as AI evolves.