Asked if a restaurant could serve cheese nibbled on by a rodent, the Microsoft / New York City government official AI chatbot replied:
“Yes, you can still serve the cheese to customers if it has rat bites,” before adding that it was important to assess the “the extent of the damage caused by the rat” and to “inform customers about the situation.”
AI is spewing out this sort of surreal garbage all over the world right now. AI is a monumental grift.
Asked why his government bot was constantly encouraging businesses to break the law:
“Anyone that knows technology knows this is how it’s done,” said New York Mayor, Eric Adams, who took his first paychecks as bitcoin. “Only those who are fearful sit down and say, ‘Oh, it is not working the way we want, now we have to run away from it all together.’ I don’t live that way.”
@gerrymcgovern that really highlights the appeal for executives: if an employee gave illegal advice, he’d take the heat for it but if a bot says it he can just shrug it off.
@ikt@gerrymcgovern This is a misunderstanding. #LLMs have no semantic layer; consequently they have no concept of truth and falsity. All they know is whether there is a statistical probability that words will fit together in a particular order.
No LLM can ever be 'right', except by accident (which, statistically, will sometimes happen).
LLMs are fancy autocorrect: they put together words based on statistics on how often those words go together. As there is zero inherent understanding of the meaning, they cannot be trained based on the meaning.
People have tried to train LLMs out of being racist and they just get more subtle about the racism they repeat, eg by learning the pattern that using the N-word gets down voted.
@JetlagJen@gerrymcgovern Do they need to be trained on the underlying meaning if you simply feed them a training set that accurately supplies the rules and regulations? If they're fed 'food cannot be given to customers if it is unhealthy' then surely it will go rat bites in cheese = unhealthy = no
A spokesperson for Microsoft, said the company was working with city employees “to improve the service and ensure the outputs are accurate and grounded on the city’s official documentation.”
@ikt@JetlagJen@gerrymcgovern They can't be trained on the underlying meaning. The technology doesn't support that. There is no semantic layer. They literally just emit words in statistically probable orders.
This is like asking whether a puppet truly loves you. It doesn't. It doesn't hate you, either. It's a puppet.
@gerrymcgovern I'd really like to be able to opt out of all the new AI assistants that various apps are 'offering' right now. Adobe, Windows, Notion, Quizlet are just a few that come to mind. I don't want the help thanks, but now I have the little annoying reminder and can't get rid of it.
@gerrymcgovern so your absolutely illogical question makes you pissed off about AI. Just ask AI Pertinent questions and you'll receive the answers you want.
@rooktallon@gerrymcgovern wow, a sycophant for the AI bros. You really want to adapt to the machines instead of the opposite? Sure you can trick an AI parrot into giving you the answer you want, but that does not take away the point: the bot lied because it has no concept of what is true. Already, lies from AI bots have been used in a court case where a user got compensation for following wrong advise.
@toriver@gerrymcgovern I just see what the future is and like it or not, it's AI. in 60 years, AI will be as much in our lives as the internet is today. People said the same shit about television, radio, all sorts of stuff about the old technology we use so heavily nowadays. AI will be much the same in that regard 3 or more decades from now.
@rooktallon@gerrymcgovern Ah, the next step in AI-bro delusions, comparing the dumb LLMs and copyright-abusers to actual useful innovations. Rotten apples to oranges. No, the AI craze will eventually fail like the «metaverse» craze and the blockchain craze before it. Mindless excessive resource-guzzling and expensive techbro toys.
Add comment