Hacker News Reader

Comment by sailingparrot | Hacker News Reader

This makes no sense. This would be like being outraged that people working in customer service are not allowed to say anything racist, or that the autocomplete in Google will never autocomplete "kill all" with a particular group of people.

Because that's what GPT-3 is used for, it's not supposed to be an unbiased reflection of society or whatever you think it is.

View on HN

> that the autocomplete in Google will never autocomplete "kill all" with a particular group of people

except "men", which it does currently.

Wait, let me check my privilege...

> that the autocomplete in Google will never autocomplete "kill all" with a particular group of people

For whatever reason, for me the #2 autocomplete in Google for "kill all" is "artists".

I'm 100% a programmer and google knows it. Mine is kill all processes by name.

I hate javascript, but mine is "kill all node processes". Never touched node with a ten foot pole. Google looks like using GPT-2.

There are already protocols to control search results and the behaviors of employees. This is very different. This is controlling the discourse of everyday people. Just because the technology is not currently deployed to restrict topics that effect you does not mean that it never will. The post above is right. Extremism does not fester in “safe spaces”. One day you may not be able to even talk about <redacted>

> This is controlling the discourse of everyday people.

I think you might be confused about the subject here.

Just responding to conversation above. I think this is important. What makes the HN interesting is sharing ideas. This is otherwise a boring corporate announcement.

The "conversation above" was

> Because that's what GPT-3 is used for

not "the discourse of everyday people"

> What makes the HN interesting is sharing ideas

Which at least occasionally have to be new ideas. If every conversation ends up on free speech absolutism, even when there literally isn't a person generating the speech, there aren't any new ideas being shared.

> Extremism does not fester in “safe spaces”.

Disagree. "Safe spaces" are bubbles of like-thinkers, else they wouldn't be "safe". People in such bubbles tend to drift to extremes due to the lack of external constraint. Within the bubble any other opinions seem extreme.

Yes, I agree. My point (and that of one of the comments earlier) is that censorship does not actually do anything to reduce extremism in the real world, and can actually increase it, as you say. So these echo-chambers are bad, which we can agree on.

> "This is very different. This is controlling the discourse of everyday people."

GPT-3 is not an "everyday people", we don't have sentient AI and laws making robots defacto people just quite yet

Reply on news.ycombinator.com