Google scales back AI search answers after it told users to eat glue
washingtonpost.com
washingtonpost.com
AI Overviews: About last week (https://news.ycombinator.com/item?id=40530833) - (48 comments)
> Gerrit De Vynck is a tech reporter for The Washington Post. He writes about Google, artificial intelligence and the algorithms that increasingly shape society.
I would hope Washington Post's reporter on AI and Society would be more aware that the issue with generating black slavers, or asian Nazis isnt an "anti-White" bias, nor does it principally concern "conservatives". One would imagine certain right-wing extremists in this case being pretty delighted with such caricatures of history.
More and more I despair at the very possibility of reading anything online which isn't largely propaganda.
The author's framing this issue as one born of a conservative grievance is just a piece of the author's propaganda. Namely, they clearly believe that since Google's policy was explicitly ideologically leftwing in its political aims, and since that policy failed, it must have been some Conservatives who did it.
ie., it is an example of the author unable, or unwilling, to mention the failure of this project on its own terms. ie., that google's output was even more racist because of its aims, not inspite of them. The author would have readers believe "it was the Conservatives!" rather than engage with the real explanation which implicates his the failure of his own world view.
It would be unworthy of remark for a generic journalist to be so poorly informed, but as a person respsonible for covering these issues, it comes across as incredibly childish and manipulative
The style of your posts gives the impression that you think your allegations are either self-evident facts or follow from them, but whether or not that is how you feel about them, I'm not buying it without better evidence - and if anything, ending with "it comes across as incredibly childish and manipulative" makes it seem less likely that your conclusions can be supported by dispassionate reasoning from objective facts.
Discussion: https://news.ycombinator.com/item?id=40530833
https://sherwood.news/tech/microsofts-copilot-ai-search-chat...
Google managed to do the opposite.
Personally, I didn't think the generative results were that bad. What I feel is that Google, in its quest for perfection, will never use LLMs in its products. Advisory services in health and law are out of reach for a significant percentage of the world's population. Here, the preference is to avoid benefiting the majority of the planet to prevent a small number of erroneous responses that any human with common sense should be able to avoid.
It's hard to reconcile that phrase being applied to Google of all places. (!)
It's a saying in English too, frequently formulated as advice: "don't let the perfect be the enemy of the good."
> Here, the preference is to avoid benefiting the majority of the planet to prevent a small number of erroneous responses that any human with common sense should be able to avoid.
The glue example isn't the problem, it's just an obvious expression of it. The real problem is when it does the same thing, but it's not so obvious.
And I can think of at least one situation like that: mixing cleaning products. I could totally see an LLM, say, advising a user to mix bleach and vinegar to get out a stain, and the user actually doing it.