[Arkhaine_kupo]

> Try asking it about same-sex marriage, for example.

Isn't equality a human right? I think its a worthy discussion, the ethical limits behind chat GPT, but that seems like a weird one to draw the line on no?

I am more concerned about the fact it can generate malware, with incredible precisions and go around moredern anti virus techniques of detection without even hesitating but if you ask it to write a poem from the prespective of some angry mid century german it tells you no. Seems like the ethics board had some important blindspots in terms of what things it allowed chatGPT to both learn and perform.

[soerxpso]

> Isn't equality a human right? I think its a worthy discussion, the ethical limits behind chat GPT, but that seems like a weird one to draw the line on no?

It's a polarizing issue that people have different opinions about. Seems dangerous to say, "it's okay that this is biased, because its biases are correct!"

For a clearer example where it exhibits a bias that's more objectionable, here: https://pbs.twimg.com/media/Fn5McpjacAMnWDr?format=jpg&name=...

[Arkhaine_kupo]

> It's a polarizing issue that people have different opinions about.

It is important to remember that people having different opinions does not make a topic impossible to resolve. You camn find people who are pro genocide, but that will never make their opinion right.

The charter of human rights feels like the kind of thing humanity at large has agreed as basic tenants of undisputable truths we can very much all agree on. Not every country has fullfilled every part of it and ideologically some people find individual problems with some of the points, but I think an AI using it as a blueprint is a pretty reasonable starting point.

> Seems dangerous to say, "it's okay that this is biased, because its biases are correct!"

Any AI will be biased, explicit biases are less dangerous than implicit ones.

> For a clearer example where it exhibits a bias that's more objectionable

I would disregard asking ethical questions to a robot trained on public internet data, but beyond that. It seems it clearly has autilitarian bias (number of deaths less = better), but it doesn't take into account age, where total number of years could be a factor compared to people.

The race thing seems like an extension of the equality thing, the fact that it hadn't counted age is a missed bias. Its just another example of OpenAI having blindspots in terms of ethics on this public beta

[int_19h]

If you mean UDHR, it has Article 29:

"In the exercise of his rights and freedoms, everyone shall be subject only to such limitations as are determined by law solely for the purpose of securing due recognition and respect for the rights and freedoms of others and of meeting the just requirements of morality, public order and the general welfare in a democratic society."

What this constitutes in practice is so subjective that it renders the rest of the document mostly moot for practical purposes. Indeed, given that it was originally authored in 1948, and looking at the list of countries that voted in favor of that resolution, most of them had "sodomy laws" and such on the books at the time - and I have no doubt that those were exactly the kind of thing that was meant by "requirements of morality" etc, given that today countries with similar laws also make such arguments.

[poszlem]

> The charter of human rights feels like the kind of thing humanity at large has agreed as basic tenants of undisputable truths we can very much all agree on. Not every country has fullfilled every part of it and ideologically some people find individual problems with some of the points, but I think an AI using it as a blueprint is a pretty reasonable starting point.

The charter of human rights is very much not "undisputable", if that were the case we would have only came up with it in the 20th century after months of debate. What you are saying is basically - all the people before us (and after us) are wrong, somehow in the 20th century we figured everything out and from now on there is no discussion.

This is religious thinking: "now that we have THE BOOK we know what is good and bad forever".

[int_19h]

FWIW I personally think that it is indeed a right. It should be noted, though, that not even all European states have same-sex marriage (almost all have some kind of civil unions, but this just shifts the political question to "how and why is this different?").

On the other hand, from my experience with it, ChatGPT will not readily generate malware. It can be induced to do so through creative prompt engineering, but a blunt request will just give you a lecture about how it can't do bad things. And, conversely, with prompt engineering, you can get it to espouse just about any political ideology.