Hacker News Reader

Comment by Cpoll | Hacker News Reader

Ironically, it seems to have parsed the challenge out from an article about the dangers of such a challenge: https://ourcommunitynow.com/news-national/watch-out-parentst....

View on HN

Reminds me of Google instructions for treating a seizure, flipping "do nots" into "do's": https://nitter.net/soft/status/1449406390976409600

I wish stuff like that would convince most people to actually read the pages' contents instead of consuming Google's digested excrement. I hate how Google Search is trying to turn into something other than a search engine.

It seems to get stuff wrong as often as not...

I bet that is confirmation bias. You probably don't notice all the times it gets it right, but really notice when it gets it wrong.

Google gets search results so hilariously (or horrifically) wrong so often that I stopped using it, ever. If ddg doesn't have what I need, I'll ask on matrix or IRC or literally anywhere else.

Google removed nuance from search terms and results a long time ago, and sometime after 2014 took to flat out lying about when pages were first seen. This makes it completely worthless for research, education, and basically anything. I was going to say looking up a phone number for a business, but Google maps is the faster way to get a valid result for that, as it stands.

So, after spending a decade and a half becoming the Kleenex of search engines, they made themselves just as disposable, for the name of ad revenue.

Good work, everyone!

> for the name of ad revenue.

I mean, that is literally why the exist? They only provide a search service in order to generate ad revenue… if they didn’t have ad revenue, they would just shut down search.

when they started they were competing against yahoo and altavista and others which had less useful results. Their service has akways been about giving good results. Now that they have results arguably worse than 1998 search engines, their ads aren't going to pay the bills for much longer. In fact it already seems to be that Google ads aren't creating as much revenue as it used to for advertisers.

Google's featured snippets is a waste of screen real estate and time.

In an emergency situation you can't blame people for not clicking through, because of ads, popups, etc. which all take valuable time to ignore/load/get past.

You make yet another excellent argument against the insane proliferation of out of control advertising to add to the already huge list of reasons that human society needs to bring that garbage industry under control.

Google also once presented an official canada.ca website as claiming you didn't need to pay income tax, because it had been declared unconstitutional in a court case: https://mobile.twitter.com/cldellow/status/13174644326168985...

What I find even more worrying than the wrong info box itself is that Google still gives the wrong answer to the "Central park value" search query from the start of that twitter thread, more than a year later and after countless people have probably reported the info box as problematic. You would think they would at least have a manual human review system.

It's not "ironic", it's one more proof that "AI" doesn't work. The danger of AI is not that it will take over the world by itself, it's that we will trust it to drive us off cliffs.

Shorting a penny across mains is not that dangerous. It’s only dangerous if the holder of the penny is grounded.

By providing AI this feedback, we can teach it to be less like the three year old that it is.

How does feedback work in this case. The AI (a misnomer) does not feel pain.

The feedback is provided as a training case for the AI. While language AIs may not actually feel pain, they can understand the danger in terms of language. This requires a data store to remember the conversations it is having, which is something that isn't widely implemented in AI.

In this case, the focus is on the idea of an AI suggesting something irrational and then providing proper feedback for how it can understand recommending similar things in the future.

Having a polarized conversation about this on a social link sharing site isn't going to help the AI understand this, which is the point I was making.

It will cause a large current to flow, often resulting in fire

This isn't proof that AI "doesn't work" any more than plane crashes are proof that air travel doesn't work.

I think that's why YouTubers self-censor the word "suicide". Intent is currently difficult to parse, so mod bots probably have a low threshold for blocking related content. I think TikTok had this problem with bomb threats to schools, claiming that the only content they found was imploring people to be aware of the threats, not actually engage in them.

Advertisers don't want to appear on videos about suicide, drugs, other topics. It's either a list their ad agency provides or one YouTube has.

The YouTubers also used to avoid the words COVID and pandemic but I think that calmed down a bit due to the sheer quantity of content that mentioned it.

Oddly, advertisers don't seem bothered that their brands are on videos where the creator says "Roll 1 twenty-sided unalive," with a massive eye-roll. I go out of my way to steer clear of the stuff that I see advertised on YouTube.

It seems "unalive" is spreading. I've heard it on several platforms now. It's pushing on meme level due to the misinterpretation by automods. There are several other words, like Heckmet instead of Helmet where it's a syllable rather than a whole word. I've even hear unalivemond for diamond

That's clbuttic. How doubleplusungood.

I'm reminded of my friend's autistic kid who loves to learn things on YouTube. Whenever he gets excited and tells you about something, he ends his spiel with "Like and subscribe!" before walking away.

These innocuous behaviors feel like a mental contagion that we should really be trying to rid ourselves of.

This is exactly what you should expect from this kind of censorship - it will simply cause the slang to cycle faster.

Stop this ride, I want to get off.

Shazould jazust shazuse dazubble dazutch.

I didn't know this was a thing on U.S. social media now... I remember reading some years ago about how Chinese social media users used homophones to bypass the automatic censorship.

https://gvu.gatech.edu/research/projects/algorithmically-byp...

https://www.amnesty.org/en/latest/news/2020/03/china-social-...

None of this is surprising. As it is today "AI" including GPT3 knows the words but not the meaning of language. AI does not understand.

Reply on news.ycombinator.com