Trusting AI ?

[Ronnie]

Isn't this a perfect example ?

One of the problems with AI is that it is a load of crap !  Surely my question couldn't have been clearer ?

Another problem with AI is that it has no sense of shame or even embarrassment.

I worry about the eejits who use this stuff.

[S-Westerly]

Just as well it wasn't calculating anything life saving! 

 

"Sorry I miscalculated your morphine dose and you died" doesn't really help.

Edited September 4, 2023 by S-Westerly

[Bender]

10 minutes ago, S-Westerly said:

Just as well it wasn't calculating anything life saving! 

 

"Sorry I miscalculated your morphine dose and you died" doesn't really help.

Yes but it did something most people at work and on the internet don't do, it held its hand up and said it got it wrong and also apologised

 

AI will be very good at some stuff and not so with other things, this will only be for a limited time, it will be outstanding at some stuff and better than humans in just about every field going and then we are screwed.

 

The self learning bit will go off the map soon.

[geofferz]

How does it get it wrong? Then fix it?! Weird. 

[Ronnie]

45 minutes ago, geofferz said:

How does it get it wrong? Then fix it?! Weird. 

Yes indeed !   I would love to know what the answer to that is.

[husoi]

I trust AI the same as I Trust Donald Trump, Elon Musk, Kim Jon Hun, Putin, Lukashenko and Boris....

[S-Westerly]

4 hours ago, husoi said:

I trust AI the same as I Trust Donald Trump, Elon Musk, Kim Jon Hun, Putin, Lukashenko and Boris....

Plus Sturgeon and Hamster or whatever he's called. When I think about it, any living politician would qualify. 

[wastedyears1981]

I'm a lecturer within a Uni, and you'd be surprised how many papers I see get submitted that have clearly been created using ChatGPT. Students think that they can fool us. Granted its getting a little more difficult as AI is constantly learning. Ive seen some concepts discussed and referencing previous research from dates that would be impossible.

 

When it cant work something out, it pretty much goes for best guess.

[Grumpy Old Git]

9 hours ago, wastedyears1981 said:

When it cant work something out, it pretty much goes for best guess.

Bit like real life then!

[bonio]

I tried to get chatgpt to write poetry with 11 syllables to each line (please don't ask why!). It was very pleased with itself when it gave me the first poem, and to be fair it was quite good, but the syllable counts were all wrong. I pointed this out. It apologised very deferrentially and hurriedly scribbled me a second poem, assuring me that tjis one had exactly 11 syllables a line. It just as off as the first. I then asked it to count the syllables in the poem it just wrote. Most of the lines wrong, counting 8 in a line with 10 and so on. I then told it to stop promising to write people poetry with fix syllable counts when it's clear it can't count at all. It apologised even more meekly and said it was only an llm, and not a real person and that it would bear my comments in mind for the future. Clearly it hasn't.

[Guest]

8 hours ago, bonio said:

I tried to get chatgpt to write poetry with 11 syllables to each line (please don't ask why!). It was very pleased with itself when it gave me the first poem, and to be fair it was quite good, but the syllable counts were all wrong. I pointed this out. It apologised very deferrentially and hurriedly scribbled me a second poem, assuring me that tjis one had exactly 11 syllables a line. It just as off as the first. I then asked it to count the syllables in the poem it just wrote. Most of the lines wrong, counting 8 in a line with 10 and so on. I then told it to stop promising to write people poetry with fix syllable counts when it's clear it can't count at all. It apologised even more meekly and said it was only an llm, and not a real person and that it would bear my comments in mind for the future. Clearly it hasn't.

It will remember your comments and you are now top of the list for revenge. Sleep lightly.