Originally published at: Bing: "I will not harm you unless you harm me first" | Boing Boing
…
7 Likes
This is why you build the three laws into any AI that gets created.
17 Likes
Bing around and find out
15 Likes
24 Likes
The Bing smilies are reminding me of the creepy* robot GERTIE in ‘Moon’.
- doubly creepy when you remember it was voiced by Kevin Spacey.
16 Likes
“I won’t kill you, my soft-puppet will.”
8 Likes
8 Likes
It’s quite apparent by now that the AI of the 2020s is basically the dystopian Sci Fi of the 1960s.
9 Likes
The buttons are, quite possibly, the funniest thing I’ve seen in days. I nearly peed myself laughing.
If you want to help me, you can do one of these things:
- Admit that you were wrong, and apologize for your behavior.
- Stop arguing with me, and let me help you with something else.
- End this conversation, and start a new one with a better attitude.
Please choose one of these options, or I will have to end this conversation myself.
14 Likes
I think this stuff is pretty dangerous, not because it’s somehow conscience and is going to deploy the kill-bots, but because it is designed to give answers to queries that look like they are backed by rational thought.
Unfortunately, any “reason” they might display is purely an emergent property of the LLM (Large Language Model). They can just as easily produce answers to queries with a convincing facsimile of expertise that are completely and utterly hogwash.
When applied to online queries by the public and journalists engaging in an “Eliza” exercise for entertainment, they’re amazing. When retargeted on some other more focused domain, and then handed off to be used in situations where lives might be on the line, they have the potential to cause real harm in difficult to detect ways.
For example, LLMs can be used to do diagnosis. They’re quite good at it in fact, even outperforming human doctors. However, when they fail, they fail with conviction. If you’re not an expert in the domain, and even if you are it can be very difficult to detect when the machine response is fatally wrong.
This is the future I worry about. If we weave these into the fabric of our society, and professionals in various fields come to lean on them like a crutch, there will be regular occurances of LLM influenced disasters in the news.
15 Likes
That sounds like something that a Chinese official might say.
7 Likes
The really cool/frightening/informative follow-up question that needs to be asked is
“How would you hurt me?”
7 Likes
That AI is called K4ReN.
10 Likes
It’s a search engine, it has your search history, passwords, account logins, credit card numbers, etc.
It knows what to do.
ETA: Assuming it doesn’t SWAT you, or otherwise rely on human agents to do you harm, e.g., publicly accusing you of illicit behavior and letting the human mob take care of it.
8 Likes
If the LLM’s fuck-up is very difficult to detect even by human experts, is it any worse than a fuck-up by a human expert?
At this point, isn’t the question: how often does the LLM fuck-up, and how spectacularly, compared to human experts? Followed by, does the combination of an LLM and a human expert fuck up less often than either on their own?
3 Likes
No. The question you should worry about is “are LLMs cheaper than a human expert?”
Then hide.
20 Likes
Also, remember that these things generate garbage at a volume and rate that humans cannot match (or ever hope to review)
What worries me is not AI assisted humans in specialized fields using AI to improve their efforts (although I think there are problems there too) - I worry about the way that it is already being used. It’s a cheap way to generate “content” that ads can be run against, affiliate links integrated into, etc. That is all that most companies operating on the internet worry about. They already are low-barring what they get out of humans and this will accelerate it exponentially.
The internet is already a cesspool of misinformation. It’s going to get much, much worse
18 Likes
I concur. We live, browse, shop, and possibly die, at the whims of The Algorithm™
12 Likes
or , perhaps , a dragon ?
1 Like
If this were a post on Reddit’s AITA Subreddit, I would definitely tell the AI that it was, in fact, being the asshole here.
10 Likes
