Why is AI so goddam awful?

[simon43]

Firstly, this is not a criticism of AN or the use of AI in their news reports!  This is a general comment about AI.

 

However, I was reading the recent news report about a British man who has been hospitalised in Thailand, with no medical insurance etc.  (You've heard this story time and time again!).

 

But what struck me was how AI news reports bear little similarity to actual human journalists who might report a similar story.

 

The vocabulary used is.. well ... basically not human! 🙂  No real person would use the vocabulary and phrases that are used in AI news reports.

 

... and that makes me think "Why?".  I thought that AI engines try to mimic humans, not robots.  Why do AI news applications use vocabulary and phrases that immediately identify them as AI?  (Perhaps that's what they are trying to do...)

 

As reported in another thread from about 2 years ago:

 

https://aseannow.com/topic/1290809-so-called-artificial-intelligence-ai/page/3/

 

19 years ago, I wrote the 'AI' SMS text-chat program called Natachata which the BBC stated was (at the time) the best candidate for passing the Turing test. http://news.bbc.co.uk/1/hi/magazine/3503465.stm

 

(in other words, fooling a real human that they were actually chatting with another human).

 

I'm sorely disappointed with the vocabulary and style of writing of the AI 'news bots'. Definitely room for improvement 🙂

[BangkokReady]

The same reason this is a valid sentence: Colorless green ideas sleep furiously

 

AI knows the rules, but it doesn't know what makes sense.

[G Rex]

22 minutes ago, simon43 said:

The vocabulary used is.. well ... basically not human! 🙂  No real person would use the vocabulary and phrases that are used in AI news reports.

 

 

My thoughts exactly!!

 

The article lost me after referring to the victims 'enigmatic disease'.  No (normal) human has ever referred to an affliction as 'enigmatic'! 

[simon43]

4 minutes ago, G Rex said:

 

My thoughts exactly!!

 

The article lost me after referring to the victims 'enigmatic disease'.  No (normal) human has ever referred to an affliction as 'enigmatic'! 

Yes, but again to reassure the mods, I'm not specifically criticising that news report or any of the news reports on AN 🙂

 

But if I ask Chat GPT to comment about something, it does seem to respond with more human-like sentences.  I guess there are different AI chat engines...

[dinsdale]

It's just all flowery rubbish that reads like a two bit novel. Sadly this is now the reality (oh the irony) that we have to put up with. 

[Tropicalevo]

20 minutes ago, dinsdale said:

It's just all flowery rubbish that reads like a two bit novel. Sadly this is now the reality (oh the irony) that we have to put up with. 

If used correctly, I understand that we do not have to put up with it.

AI does what you ask it to (in theory).

if you want flowery rubbish padding out basic statements, that is what you get.

GIGO

[Celsius]

It's not awful and it has its uses.

 

It helps Bob come up with new topic every day.

[OneMoreFarang]

1 minute ago, Celsius said:

It's not awful and it has its uses.

 

It helps Bob come up with new topic every day.

Actually, that is awful. 

[OneMoreFarang]

I am sure AI will get better over time.

They have to start somewhere and have their priorities.

I ask AI sometimes technical questions or maybe something about history. I am looking for information and not poetic language.

I guess that is what AI developer want in the moment.

Why would any AI developer spend much time to read a couple of news articles and then formulate them somehow differently? Who needs something like that?

[dinsdale]

5 minutes ago, Tropicalevo said:

If used correctly, I understand that we do not have to put up with it.

AI does what you ask it to (in theory).

if you want flowery rubbish padding out basic statements, that is what you get.

GIGO

I agree. What AI spits out is heavily dependent on the prompts.

 

[GammaGlobulin]

1 hour ago, BangkokReady said:

Colorless green ideas sleep furiously

 

Looks like somebody has been reading the early work of Chomsky.

I think this might come from his first PhD thesis, or something.

Or, his first "infamous" paper, one which he himself had later stated he did not entirely agree with.

A long time ago...

 

Syntactic Structures

As I recall.

Or, maybe like any good AI...

I am just hallucinating...

Again.

 

[GammaGlobulin]

57 minutes ago, BangkokReady said:

AI knows the rules

 

AI does not know the rules.

Neither does it know anything.

 

[brfsa2]

oh, please... it's not AI that is awful, it's how people use, it's a tool, a very powerful tool, just like a gun,

how do you use a gun? careful

how do you use AI? careless

 

that sums it up.

I create tools with AI, use it for almost most of my job, rely on it, cant be effective anymore without it.

 

then there are the people who has no clue about anything... posting here.

 

[spidermike007]

I agree AI is absolutely horrific. I'm told by friends in the military that they have regenerative AI which is on a completely different level and is infinitely more intelligent than the kind of AI that we have access to. And that scares me, and should scare most. It is nothing like the AI we see, which is spectacularly dumb. 

[Thailand]

Apparently Musk's Grok 3 is a big  step beyond what we have seen so far.

Have used Grok 2 and it is good. hoping to try Grok 3 in the next day or two. 

But it does seem quite scary with rapid growth and moving in to the unknown with Super Intelligence , Living Intelligence etc which is something else again!

[GammaGlobulin]

17 minutes ago, spidermike007 said:

I agree AI is absolutely horrific. I'm told by friends in the military that they have regenerative AI which is on a completely different level and is infinitely more intelligent than the kind of AI that we have access to. And that scares me, and should scare most. It is nothing like the AI we see, which is spectacularly dumb. 

 

It's impossible to prove that your statement is not correct.

 

[sidjameson]

In evolution the first single cell organism appeared 3.5 billion years ago.

The first multi cell organism 2 billion years ago.

 

I liken the first attempts at AI to the first single cell organism. The recent batch of AI to the first multi cell organisms.

It took 20 years or so.

 

At the same rate of progress AI in 25 years will be to it's current form what humans are sino bacteria.

[ukrules]

1 hour ago, spidermike007 said:

I agree AI is absolutely horrific. I'm told by friends in the military that they have regenerative AI which is on a completely different level and is infinitely more intelligent than the kind of AI that we have access to. And that scares me, and should scare most. It is nothing like the AI we see, which is spectacularly dumb. 

 

Have you used any of the higher end paid for subscription models?

[cdemundo]

Even simple automation is not very reliable.

I use the self checkout at a local supermarket 5 or 6 times a week.

At least once a week it has some kind of error that requires a store employee to intervene with a bypass key.

Not as frequently all four of the self checkout machines will have a customer waiting for assistance with a locked up machine, and I will go to a human checkout.

 

I am not posting to complain about the inconvenience, rather I offer an example of relatively simple automation and its lack of reliability.  What level of reliability can we expect from advanced technology?

[NorthernRyland]

4 hours ago, simon43 said:

But what struck me was how AI news reports bear little similarity to actual human journalists who might report a similar story.

 

 

I've read enough of these to tell you they are not replacements for humans and will trash the quality of the news on this site. I'm already sick of it. It omits critical details that anyone would want to know and frustrates readers. Companies that rely on this for news will lose readership I'm sure of it.

 

It doesn't help they'll delete comments and give bans when people point this out. It's fair criticism and the readers are the customers so I fail to see how this is a good policy long term.

[NorthernRyland]

4 hours ago, dinsdale said:

It's just all flowery rubbish that reads like a two bit novel.

It's an insult to the readers. Makes you feel like a muppet reading such things about serious topics. I skip right to the comments and hear what real people have to say instead.

[connda]

Want to know how come AI is being used for journalism?  Because humans hired by the MSM can't do any better.  So why not.  If you want actual journalism, turn off the MSM and go 100% independent media. 

[tomazbodner]

Because AI isn't actually any form of intelligence. It is a large model, which was trained by simply rating the randomly produced output for likeability. It has no idea what it means, but it seems people like it.

[madone]

6 hours ago, G Rex said:

 

My thoughts exactly!!

 

The article lost me after referring to the victims 'enigmatic disease'.  No (normal) human has ever referred to an affliction as 'enigmatic'! 

 

nonsense -- the article is crap of a million reasons, but a disease no one can identify is enigmatic. It is a word commonly used in medical circles

 https://www.google.com/search?q=enigmatic+diesase&oq=en&gs

 

apologies the site wont accept the google link

[Pouatchee]

10 hours ago, BangkokReady said:

The same reason this is a valid sentence: Colorless green ideas sleep furiously

 

AI knows the rules, but it doesn't know what makes sense.

 

wasnt it noam chomsky who said that?

[Pouatchee]

10 hours ago, simon43 said:

But if I ask Chat GPT to comment about something, it does seem to respond with more human-like sentences.  I guess there are different AI chat engines...

 

i pay for chatgpt and overall enjoy it a lot even if it makes a lot of mistakes. This weekend i asked it to create a picture of a teacher whose head is exploding under pressure. I didn't like the mustaches it kept putting in. I kept telling it to edit the pics and get rid of the mustaches. every time it generated a new photo it kept putting on a mustache telling me it had gotten rid of it. there was no way i could get it to fic it the right way... still a few clinks to iron out.. but good fun and no limitations with the paid subscription. I can afford the 28$ US a month so no biggy..

[simon43]

3 hours ago, Pouatchee said:

 

i pay for chatgpt and overall enjoy it a lot even if it makes a lot of mistakes. This weekend i asked it to create a picture of a teacher whose head is exploding under pressure. I didn't like the mustaches it kept putting in. I kept telling it to edit the pics and get rid of the mustaches. every time it generated a new photo it kept putting on a mustache telling me it had gotten rid of it. there was no way i could get it to fic it the right way... still a few clinks to iron out.. but good fun and no limitations with the paid subscription. I can afford the 28$ US a month so no biggy..

Perhaps ChatGPT was written by a Brit and it can't understand what you mean by 'mustache' 🙂

[captain_shane]

13 hours ago, spidermike007 said:

I agree AI is absolutely horrific. I'm told by friends in the military that they have regenerative AI which is on a completely different level and is infinitely more intelligent than the kind of AI that we have access to. And that scares me, and should scare most. It is nothing like the AI we see, which is spectacularly dumb. 

Your friends are lying. Did the government somehow secretly create a cluster of millions of advanced gpu's?

[it is what it is]

 

AI is amazing. however, as with much technology, garbage in, garbage out. i suspect the people using AI often don't really understand how it works, or are too lazy to do more then push a button and then publish, with no QC.

 

I use AI in my writing work and photography, used correctly it's an absolute game changer.

[CharlieH]

Actually, in my experience, it depends entirely on the "prompt" used. There is a whole industry evolved in "prompt engineering".

 

If you dont tell it what you want, you will get default setting responses. tell it to phrase or write a certain way and it will do that, well, almost. It isn't perfect, agreed, but it can be manipulated quite a bit.