b) Society is already plenty de-sensitised to violence, sex and whatever other horrors anyone has conceived of in the last century of content production. There is nothing an LLM can come up with that has or is going to shock anyone.
c) The most popular use cases for these unleashed models seems to be as expected deepfakes of high school girls by their peers. Nothing that is moving society forward.
OpenAI "moves society forward," Microsoft "moves society forward." I'm sincerely uninterested in progress, it always seems like progress just so happens to be very enriching for those who claim it.
>There are plenty of models out there without guard rails.
Not being used at a mass scale.
>Society is already plenty de-sensitised to violence, sex and whatever other horrors anyone has conceived of in the last century of content production. There is nothing an LLM can come up with that has or is going to shock anyone.
Oh, but it wouldn't really be very shocking if you could expect it, now would it?
Just that we had unleashed models for a while now and the only popular use case for them has been deep fakes. Otherwise it's just boring, generic content that is no different to what you find on X or 4chan. It's 2024 not 1924 - the world has already seen every horror imaginable many times over.
And not sure why you think if they were mass scale it would change anything. Most of the world prefers moderated products and services.
Yes, the very same "moderated" products and services that have risen sea surface temperatures so high that at least 3 category 4+ hurricanes, 5 major wildfires, and at least one potential or actual pandemic spreads unabated every year. Oh, but don't worry, they won't let everyone die: then there would be no one to buy their "products and services."
Although I do agree with you that in this particular situation, the LLM safety features have often felt unnecessary, especially because my primary use case for ChatGPT is asking critical questions about history. When it comes to history, every LLM seems to have an increasingly robust guardrail against making any sort of definitive statement, even after it presents a wealth of supporting evidence.
Hundreds of thousands of people are making AI porn in their basements and deleting 99.99% of it when they are… finished.
Hundreds of people are making deep fakes of people they know in some public forums.
And, how does the public interpret all of this?
“The most popular use case is deepfake porn.”
Sigh…
Is there proof that the self censoring only affects whatever the censors intend to censor? These are neural networks, not something explainable and predictable.
That in addition to the obvious problem of who decides what to censor.
b) No, certain things aren't taboo anymore, but new taboos emerged. Watch a few older movies and count "wow this wouldn't fly nowadays" moments
c) This was exactly the same use case the internet had back when it was fun, and full of creativity.