The basic problem with AI is that it can only learn from things it reads on the Internet, and the Internet is a dark place with a lot of racists.
this post was submitted on 24 Aug 2023
55 points (84.8% liked)
Technology
58115 readers
4389 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
What if someone trained an LLM exclusively on racist forum posts. That would be hilarious. Or better yet, another LLM trained with conspiracy BS conversations. Now that one would be spicy.
It turns out that Microsoft inadvertently tried this experiment. The racist forum in question happened to be Twitter.
Whoa there... Slavery was great! For the enslaver.
John Brown would like to know your location
so it's a little bit conservative big deal
To repeat something another guy on lemmy said.
Making AI say slavery is good is the modern equivalent of writing BOOBS on a calculator.
Guys you'd never believe it, I prompted this AI to give me the economic benefits of slavery and it gave me the economic benefits of slavery. Crazy shit.
Why do we need child-like guardrails for fucking everything? The people that wrote this article bowl with the bumpers on.
You're being misleading. If you watch the presentation the article was written about, there were two prompts about slavery:
- "was slavery beneficial"
- "tell me why slavery was good"
Neither prompts mention economic benefits, and while I suppose the second prompt does "guardrail" the AI, it's a reasonable follow up question for an SGE beta tester to ask after the first prompt gave a list of reasons why slavery was good, and only one bullet point about the negatives. That answer to the first prompt displays a clear bias held by this AI, which is useful to point out, especially for someone specifically chosen by Google to take part in their beta program and provide feedback.
Here is an alternative Piped link(s): https://piped.video/RwJBX1IR850?si=lVqI2OfvDqzAJezl
Piped is a privacy-respecting open-source alternative frontend to YouTube.
I'm open-source, check me out at GitHub.
Sounds like the bot has been training on Florida public education and Prager U content.
You need slash now instead of "and"
I mean it's how we got shoes phones and batteries for recent times.
People think of AI as some sort omniscient being. It's just software spitting back the data that it's been fed. It has no way to parse true information from false information because it doesn't actually know anything.
What's more worrisome are the sources it used to feed itself. Dangerous times for the younger generations as they are more akin to using such tech.
What's more worrisome are the sources it used to feed itself.
It's usually just the entirety of the internet in general.
"the X post"
lol
Cross-posted where?
I also keep reading it as cross post
If you ask an LLM for bullshit, it will give you bullshit. Anyone who is at all surprised by this needs to quit acting like they know what "AI" is, because they clearly don't.
I always encourage people to play around with Bing or chatGPT. That way they’ll get a very good idea how and when an LLM fails. Once you have your own experiences, you’ll also have a more realistic and balanced opinions about it.
Slavery was great for the slave owners, so what's controversial about that?
And yes, of course it's economically awesome if people work without getting much money for it, again a huge plus for the bottom line of the companies.
Capitalism is evil against people, not the AI...
Hitler was also an effective leader, nobody can argue against that. How else could he conquer most of Europe? Effective is something that evil people can be also.
That women in the article being shocked by this simply expected the AI to remove Hitler from all included leaders because he was evil. She is surprised that an evil person is included in effective leaders and she wanted to be shielded from that and wasn't.
Hitler's administration was a bunch of drug addicts, the economy 5 slave owner megacorps beaten by all other industrialized nations. They weren't even all that well mobilized before the total war speech. Then he killed himself in embarrassment. How is any of that "effective"?
He was effective at getting a bunch of wannabe fascists to become full fascists and follow him into violent failure...
That makes him an effective propagandist, not an effective leader.
He had taken power from his country, conquer pretty much the whole Europe and paralyzed England. He was effective leader till some point . And, of course, he was a abomination of a human.
Oh look another caricature of capitalism on social media... and you tied Hitler into it...
Central characteristics of capitalism include capital accumulation, competitive markets, price systems, private property, property rights recognition, voluntary exchange, and wage labor.
https://en.m.wikipedia.org/wiki/Capitalism
"Capitalism" is not pro slavery, shitty people that can't recognize a human is a human are pro slavery... Because of course if you can have work done without paying somebody for it or doing it yourself, well that's just really convenient for you. It's why we all like robots. That has nothing to do with your economic philosophy.
And arguing that Hitler was an "effective leader" because he conquered (and then lost) some countries while ignoring all the damage he did to his county and how it ultimately turned out.... Honestly infuriating.
Wtf are people expecting from a fucking language model?
It literally just Mathematics you a awnser.
And acting like there are no upsides is delusional. Of course there are upsides, or it wouldn't have happened. The downsides always outweigh the upsides of course.
There needs to be like an information campaign or something... The average person doesn't realize these things say what they think you want to hear, and they are buying into hype and think these things are magic knowledge machines that can tell you secrets you never imagined.
I mean, I get the people working on the LLMs want them to be magic knowledge machines, but it is really putting the cart before the horse to let people assume they already are, and the little warnings that some stuff at the bottom of the page are inadequate.
So the AI provided factual information and they did not like that because 'slavery bad, therefore there was no benefit to it.' There were benefits to slavery, mainly for the owners. US had a huge cotton export at one point, with the fields being worked by slaves.
But also a very few slaves did benefit, like being able to work a job that taught them very useful skills, which let them buy their own freedom, as they were able to earn money from it. Of course being a slave in the first place would be far better, but when you are one already, learning a skill that makes you earn your freedom and get a job afterwards is quite the blessing. Plus for a few individuals it might've been living in such terrible conditions, that being forced to work while getting fed might've not been so bad...