3 ms·

I saw the word "safety" a few places in the article. What does "safety" mean in this context?

i_like_apis 4y ago
In this context, it’s buzz-killing, pearl-clutching, often woke, nonsense: … attempting to limit the generation space to omit porn, copyright infringement, violence, racially “unbalanced” content … etc.
ttul 4y ago
You can ask the AI to generate a picture of horrid things, and it will oblige.
- emikulic 4y ago
  You can generate things the author doesn't like. And since you're doing it on your video card at home, nobody can stop you.
  - Buttons840 4y ago
    If a computer model could produce the world's best porn, would that be a good or bad thing? Many harmful effects of porn would be amplified, but it would reduce the exploitation of real people in the industry. A moral question society will soon face I think.
    - acdha 4y ago
      I think the bad things would be targeted use of realistic images: for example, imagine the horrible things some people experience in school multiplied by “leaked” photos, or someone’s abusive ex distributing “proof” of their infidelity / unsuitability to have custody, etc. There’s a theory that over time people would stop believing everything they see but there’s still plenty of time for millions of tragedies before that happens, if it ever does. Forensics is going to be a growth industry.
  - truffdog 4y ago
    I really don't understand why people are pretending the licensing makes any difference here.
- astrange 4y ago
  I, meanwhile, find AI research distasteful because it’s written in Python.
  - forgingahead 4y ago
    Yeah we need Trigger Warnings on any ML repo that uses Python. The trauma that comes from dealing with Python dependency management is really hurtful and non-inclusive.
- Gigachad 4y ago
  Which I don't think is the end of the world. I can draw horrid things on my iPad in Procreate or even just a pencil on paper. What's new here is ease of access and hyper realism. This is more a problem for fake news than just generating bad/shock images which were already possible.

In this context, it’s buzz-killing, pearl-clutching, often woke, nonsense: … attempting to limit the generation space to omit porn, copyright infringement, violence, racially “unbalanced” content … etc.

ttul 4y ago

You can ask the AI to generate a picture of horrid things, and it will oblige.

emikulic 4y ago
You can generate things the author doesn't like. And since you're doing it on your video card at home, nobody can stop you.
- Buttons840 4y ago
  If a computer model could produce the world's best porn, would that be a good or bad thing? Many harmful effects of porn would be amplified, but it would reduce the exploitation of real people in the industry. A moral question society will soon face I think.
  - acdha 4y ago
    I think the bad things would be targeted use of realistic images: for example, imagine the horrible things some people experience in school multiplied by “leaked” photos, or someone’s abusive ex distributing “proof” of their infidelity / unsuitability to have custody, etc. There’s a theory that over time people would stop believing everything they see but there’s still plenty of time for millions of tragedies before that happens, if it ever does. Forensics is going to be a growth industry.
- truffdog 4y ago
  I really don't understand why people are pretending the licensing makes any difference here.
astrange 4y ago
I, meanwhile, find AI research distasteful because it’s written in Python.
- forgingahead 4y ago
  Yeah we need Trigger Warnings on any ML repo that uses Python. The trauma that comes from dealing with Python dependency management is really hurtful and non-inclusive.
Gigachad 4y ago
Which I don't think is the end of the world. I can draw horrid things on my iPad in Procreate or even just a pencil on paper. What's new here is ease of access and hyper realism. This is more a problem for fake news than just generating bad/shock images which were already possible.