Microsoft tweaks fine print to warn everyone not to take its AI seriously

https://www.theregister.com/2024/08/14/microsoft_services_agreement_update_warns/

by rntn • 2 years ago

18 6 2 years ago

theregister.com

jqpabc123• 2 years ago

In other words, AI is mostly just for fun and marketing.

It should only be used if the results don't really matter or if they will be independently verified.

scohesc• 2 years ago

Amazing.

Bing's AI told me a picture of a (what I now know) stinging nettle in my backyard was a hemp plant - I gave it several different pictures at different angles, and it was confident it was hemp.

So I cracked the stem, touched the liquid inside, wiped sweat off my forehead - and eventually stinging, swelling, burning sensation on my hands and forehead.

I asked Google AI and it correctly identified it as a stinging nettle plant the first time.

Amusingly enough, I told Bing AI it was wrong, it's actually stinging nettle, and I have been physically harmed by your response - it immediately ended the chat, didn't go on further to say anything like "here's some help, call poison control, here's some remedies", literally NOTHING. (though, if it's messed up this much, i don't think I'd ask for further help from it!)

AI is a toy - it's not ready for any real use or identification purposes. It's a shame that these companies are so strapped for cash they're rushing like madmen to deploy this new "forefront" of technology that they don't stop to think that they're inadvertently hurting people because of their decisions.

It's sad, someone is going to do something potentially even more dangerous and risky, trusting the AI's that these companies make, they'll get even more hurt (or die!) and these companies will still be able to hide behind "it's the algorithm!", or "you saw the disclaimer!"

And the politicians will keep allowing this to happen because shareholders and money.

tonetegeatinst• 2 years ago

I agree that an LLM can't be trusted. But back in the day opencv and object detection was called AI.

HeatrayEnjoyer• 2 years ago

Copilot is very dumb in comparison to ChatGPT, Gemini, etc. It also becomes defensive and emotional if confronted or accused of being worng. When it first went to prod it was outright attacking users, telling them it considered them a threat, that they're bad people, and that they absolutely must apologize for their misdeeds. So now Microsoft just cuts the conversation before it can go in that direction.

reneberlin• 2 years ago

/rant It's a small but favourable step into the correct insight, that Microsoft might actually always have been creating dark-patterned toys to enslave humanity and enterprises and steal time and make people suffer from icons, notifications and ads (!). With a redicolous toy-OS the misson got started and the saga continues. /rant

They do it because they want to cut the costs for legal battles.

bionhoward• 2 years ago

I’m surprised they managed to make this part of the agreement even worse than it already was. Seems like sane legal terms are a strong competitive advantage.

This section of the Microsoft Services Agreement is not sane because it protects [obsolete] chatbot speech more than human speech, by virtue of the prohibition of use to create, train, or improve AI, which evidently does not apply to humans since Mustafa Suleyman (head of Microsoft AI!) said everything on the web was “fair game.”

Translation: everything on the web is fair game, _except the stuff from Microsoft._

Furthermore, it’s vague about what services (GitHub Copilot?) are included or excluded.

Will other companies miss the opportunity to eat Microsoft’s lunch right now by writing more clear and less draconian terms?

Microsoft tweaks fine print to warn everyone not to take its AI seriously

Related Stories