> I personally hate it when SV moral arbiters nanny me.
I've been annoyed with them pushing their mores on the global internet since at least 2010.
I suspect you'd also hate the substantially different mores that I would have in their place.
> Who is driving this big push for "safety", anyway? Do consumers actually want safety or are a concern-trolling vocal minority pressuring AI corporations to kowtow?
1. Yudkowsky, whose general vibes are an important part of the discussion for about half the people who work on these AI in the first place, even where they disagree in particulars.
2. Anyone who noticed the way biases in training data propagate stereotypes, an observation which substantially predates any of the currently interesting generators.
3. Anyone who has been on the receiving end normal old fashioned inappropriate content, or who is the parent or guardian of such a person.
4. Also the usual concern trolling types, as some people have already been arrested for using such models to sexualise specific people including, indeed, at least one case where it was a minor.
5. Anyone who can see the potential for these models in automated personalised propaganda.
6. Anyone concerned with the potential for a fully automated system that A/B tests with a constant stream of newly generated output until it finds a super-stimulus you can't help but engage with.
These groups don't all talk to each other, and in many cases dismiss the severity, likelihood, and timescales of each other, though often still using overlapping language that makes any conversations on these issues even more difficult than figuring out exactly what someone who just used "woke" as a pejorative is actually objecting to.
I've been annoyed with them pushing their mores on the global internet since at least 2010.
I suspect you'd also hate the substantially different mores that I would have in their place.
> Who is driving this big push for "safety", anyway? Do consumers actually want safety or are a concern-trolling vocal minority pressuring AI corporations to kowtow?
1. Yudkowsky, whose general vibes are an important part of the discussion for about half the people who work on these AI in the first place, even where they disagree in particulars.
2. Anyone who noticed the way biases in training data propagate stereotypes, an observation which substantially predates any of the currently interesting generators.
3. Anyone who has been on the receiving end normal old fashioned inappropriate content, or who is the parent or guardian of such a person.
4. Also the usual concern trolling types, as some people have already been arrested for using such models to sexualise specific people including, indeed, at least one case where it was a minor.
5. Anyone who can see the potential for these models in automated personalised propaganda.
6. Anyone concerned with the potential for a fully automated system that A/B tests with a constant stream of newly generated output until it finds a super-stimulus you can't help but engage with.
These groups don't all talk to each other, and in many cases dismiss the severity, likelihood, and timescales of each other, though often still using overlapping language that makes any conversations on these issues even more difficult than figuring out exactly what someone who just used "woke" as a pejorative is actually objecting to.