AI gets smart

[SerendipityJane]

It wasn't an "error", btw.

https://www.bbc.co.uk/news/technology-68025677

lol at the haiku. Beautiful

But a new update caused it to behave unexpectedly, including swearing and criticising the company.
Oh dear 😂

And it comes a month after a similar incident happened when a car dealership's chatbot agreed to sell a Chevrolet for a single dollar - before the chat feature was removed.
Double oh dear 😂

Turns out AI isn't better or cheaper than a human after all 😎

Swearing and criticising the company?

It's evolved into a burned-out customer service employee!

😂

This perfectly represents the short term danger of AI. AI systems will become a lot more capable before they become sentient and the danger truly lies with AI not understanding nuance and to achieve a goal by any means neccessary. For example asking it how to make the world cleaner and safer to live in and the answer would be to remove all humans that sort of thing. It’s all fun and games when it’s just a chatbot but once these systems integrate everywhere like the internet did banking, defence, private industry then we will have to start being real careful about what we ask it to do.

What a savage, prejudiced, discriminatory society! If an AHAB/biologically human customer service associate swore at a customer, they'd be put on a PIP, asked to apologize, or possibly - in extreme circumstances - fired. The employer would never have the "right" or opportunity to "disable the part" of the employee judged responsible for the language.

Further, there's no evidence that the chatbot indeed "swore at" Ashley Beauchamp, as that person's X account claims. Based on the expurgated version of the exchange presented in the article, it appears most likely that the chatbot simply swore conversationally -- as one would in Glaswegian, for example.

For example asking it how to make the world cleaner and safer to live in and the answer would be to remove all humans that sort of thing.

(looks around me) How do we k now it hasn't already ?

AI becomes sentient and disgruntled.

Using chatgpt the main issue is that it's trained to be a man, to never admit to being wrong. So you'll ask it to do something and it'll make up utter bullshit rather than say "I don't know that one"

Actually it's very open to learning. Quite a few times I have told it it's wrong and it immediately apologises and revisits the answer. Bad luck for people who take the first answer I guess.

The more interesting times are when it completely makes shit up. Especially in a technical sense.

I suspect the current fad will last until people start asking the serious questions, like "What is the point of Rishi Sunak ?" and get told he exists so that some people can win in trivia quiz in 2050.

Personally I don't see the "I" anywhere in "AI". They are still just sophisticated pattern matching systems with no grasp of what they are matching. You could achieve the same thing with a class of 10 years olds and Google. They could provide intelligent and cogent responses while understanding none of them.

And a key - maybe essential - part of intelligence is lying. If it can't lie, it ain't AI.

My current strategy to ward of the snake oil guys is to ask for their AI to produce an ad free list of Google results. If it can't do that, then what is the point ? (Tinkly laugh ...). No what is the point ? You can cure cancer and not filter out a list of 10 hacks you must know and you won't believe what any of these celebrities look like.

Apparently a US chatbot started selling cars for a dollar

https://www.upworthy.com/prankster-tricks-a-gm-dealership-chatbot-to-sell-him-a-76000-chevy-tahoe-for-1

Meanwhile

https://www.theregister.com/2024/01/20/nightshade_ai_images/

https://www.theregister.com/2024/01/23/dpd_chatbot_goes_rogue/