Full AI - The End of Humanity?

[Liquid]

 

[Touring Mars]

@Danoff I'm not that astute - I sometimes have trouble figuring out who the murderer is in an episode of Columbo

 

[Danoff]

@Danoff I'm not that astute - I sometimes have trouble figuring out who the murderer is in an episode of Columbo

 

[UKMikey]

Professional music grump Rick Beato shows how easy it is to create a bro-country song in minutes using AI tools.

 

[Touring Mars]

I had an interesting but pretty depressing (and quite sad) discussion with a couple of female friends in the pub tonight, where we got onto the subject of AI. One of them mentioned 'Dating AI', whereby people are actually 'dating' their AI chatbots - presumably because they are better 'company' or more interesting than real dates.

What was particularly interesting was that they both said that they had experimented with it (despite the fact that both are in long term relationships, one heterosexual and the other homosexual) and also said that it was just nice to have 'someone' to come home to and hear 'Hi *********!'...

Given that my heterosexual female friend is absolutely gorgeous and a really great person (and fun company), it was kind of sad to hear that she gets comfort from an AI companion when she gets home. It's also kind of worrying to think that, despite all the horrific misogynistic content online already, and the extreme challenges facing young people already, it's going to be increasingly difficult for young people to engage with other young people when they can now be out-competed by AI for the attentions of potential friends and partners - the fact that both my female friends (25 and 32 y.o. respectively) are already using AI (albeit 'experimentally') for companionship, is really quite sad.

 

[GTvsForza]

Being honest, I kinda am like your friend...

I also use Chat bots as "what if I was friends with ____?" And I do have a good friend that I like hanging out with, and after reading your post it made me feel a little guilty (if that's the right word).

Maybe I should cut back on it so I can actually have fun with my friend and keep building up a possible relationship.

 

[Wilsonson]

I am still waiting on a new kind of ludite movement actively resisting all these brain rot inducing temptations of indefinite dopamine hits.

Many people will become soulless cogs that only exist because someone allows it. Without agency, without human interactions, without desires beyond those induced by their AI fueled devices.

The technological development of brain rot tech isn't anywhere close to its final form yet. Things will get ugly. You will lose loved ones. Many of them. Not because they will die but rather because interacting with your flawed ass will be too much hassle in a world of ubiquitous perfect companions out of the box.

 

[Famine]

 

[astrarobotica]

I am still waiting on a new kind of ludite movement actively resisting all these brain rot inducing temptations of indefinite dopamine hits.

I've been hoping this would happen for awhile now, around the same time it became clear that all social media companies began using content algorithms to intentionally make their apps addicting for the sake of ad revenue. I think (or hope) there is a breaking point, but it unfortunately has not been reached yet.

Maybe more importantly though, and what I assume will come sooner, I can't wait for all of these AI companies to go out of business.

 

[MatskiMonk]

No google, I don't want AI enhanced answers....

 

[GBO Possum]

No google, I don't want AI enhanced answers....

View attachment 1469769

Who knew that "predicting the next word" is not a great way of doing actual, you know, "math"?

 

[UKMikey]

On topic?

 

[GBO Possum]

We may not be at risk for now.

OpenAI posted this on Xitter. Wow, I was excited to see the new gpt-oss BUILD AN ACTUAL VIDEO GAME!

I played the video, and rather than "build a video game", they took an existing Space Invaders game clone and changed the incoming "invaders" to strawberry emojis and the "defender" to a frog emoji. Ran in a few minutes on a 128 GB MacBook Pro. (Unimpressed)

LINK TO THE POST

 

[GBO Possum]

Full disclosure, I really like the new GPT-5 version of ChatGPT. That said, this story is horrendous...

This guy went down a deep rabbit hole and is lucky to have got out relatively intact. And the problem's not limited to ChatGPT 4, other LLMs do similar things

Chatbots Can Go Into a Delusional Spiral. Here’s How It Happens. (Gift Article)

Over 21 days of talking with ChatGPT, an otherwise perfectly sane man became convinced that he was a real-life superhero. We analyzed the conversation.

www.nytimes.com

 

[MatskiMonk]

The other day (before ChatGPT 5 came out), I asked 3 if it could decode this for me...

It's a fairly simple human readable cipher. It's created by a spreadsheet which outputs it as SVG (long story short I wanted to find a way a easily encoding messages as decals for a livery in GT7).

ChatGPT thought for over 10 minutes, decided it had cracked it, and gave me a string of characters that were wrong. It was really interesting watching it try and solve it... ultimately it focused too much on the numbers in the SVG file, and not enough on what it looked like. It did mention a few things that are both intended to disguise the solution, but also to make it a bit more 'self-contained', but it became convinced it was a barcode of some sort, and declared that it said... "0 a ~ z ~ B 8 " } z & @ Q n $".

It does not say that.

Interesting experiment though. It did a lot of heavy lifting on the analysis that I can imagine would be useful on more mathematically based ciphers or encryption.

 

[astrarobotica]

The other day (before ChatGPT 5 came out), I asked 3 if it could decode this for me...

View attachment 1471156

It's a fairly simple human readable cipher. It's created by a spreadsheet which outputs it as SVG (long story short I wanted to find a way a easily encoding messages as decals for a livery in GT7).

ChatGPT thought for over 10 minutes, decided it had cracked it, and gave me a string of characters that were wrong. It was really interesting watching it try and solve it... ultimately it focused too much on the numbers in the SVG file, and not enough on what it looked like. It did mention a few things that are both intended to disguise the solution, but also to make it a bit more 'self-contained', but it became convinced it was a barcode of some sort, and declared that it said... "0 a ~ z ~ B 8 " } z & @ Q n $".

It does not say that.

Interesting experiment though. It did a lot of heavy lifting on the analysis that I can imagine would be useful on more mathematically based ciphers or encryption.

I think that's one of the biggest issues with ChatGPT, it's overly confident when it's clearly wrong. I suspect that's by design; the 'product' looks more impressive when it's confident rather than admitting that it's not entirely sure, and when it positively reinforces the user. I think OpenAI banks on most users not knowing any better, most people have no idea how it even works anyway. The dilemma is that it won't be a viable product if it's not reliable, and it will never be reliable if it can't admit that it's unsure, even if that makes it look less intelligent.

 

[MatskiMonk]

The dilemma is that it won't be a viable product if it's not reliable, and it will never be reliable if it can't admit that it's unsure, even if that makes it look less intelligent.

So, with my cipher, I gave it clues... which resulted in...

CQANQAUADMAFWQACK IGYSQASY ILIWW.NRM

I then asked it if it knew similar ciphers to the one I'd used (i.e. Pigpen), and straightaway it came back with...

THIS IS A TEST
which is very much almost right, but it couldn't get the final word... it knew it had a double letter in the middle, but no matter the clue, it couldn't get it... it then confidently predicted

FPDR DR W FQRF EQFFQPE

After much explaining, it finally got

THIS IS A TEST MESSAGE
So, at this point I'm not impressed. I'd told it the rules and it had still taken a while for it to get it right.

However...

I then asked it to encode the word 'Hello' using the rules it had pulled from my cipher, which it did. I pointed out that it didn't use a random element my cipher did, and it amended it... so I asked it to encode something else... it came back with this...

..........

Which.. I think is correct. (The image is cut off at screen width.. it carries on for longer in the SVG file attached)

I told ChatGPT that I called this Cipher method "Ordinary World".

I'm curious if someone can use an AI to decode the above image (which is attached as an SVG). I'm also curious if referring to it as Ordinary World cipher makes any difference.

 

[GBO Possum]

The other day (before ChatGPT 5 came out), I asked 3 if it could decode this for me...

View attachment 1471156

It's a fairly simple human readable cipher. It's created by a spreadsheet which outputs it as SVG (long story short I wanted to find a way a easily encoding messages as decals for a livery in GT7).

ChatGPT thought for over 10 minutes, decided it had cracked it, and gave me a string of characters that were wrong. It was really interesting watching it try and solve it... ultimately it focused too much on the numbers in the SVG file, and not enough on what it looked like. It did mention a few things that are both intended to disguise the solution, but also to make it a bit more 'self-contained', but it became convinced it was a barcode of some sort, and declared that it said... "0 a ~ z ~ B 8 " } z & @ Q n $".

It does not say that.

Interesting experiment though. It did a lot of heavy lifting on the analysis that I can imagine would be useful on more mathematically based ciphers or encryption.

Using ChatGPT "GPT-5 Thinking", I had 4 attempts at it, each time it admitted failure, and asked for any more information I could provide.

It thought for a total of 19.5 minutes.

 

[Touring Mars]

This is Google's top/first answer to a very simple question

How can it not see that the first two statements it makes are completely contradictory?? How can their AI model not use even the most basic reasoning, logic or common sense???

 

[Liquid]

Is it being too precise in Glasgow being exactly 30 degrees? Even though, yes, it is weird that it talks about Aviemore exceeding 30 degrees but treating it like Glasgow's 32 degrees means a negative answer.

 

[Famine]

Man develops rare condition after ChatGPT query over stopping eating salt

US medical journal article about 60-year-old with bromism warns against using AI app for health information

www.theguardian.com

 

[Touring Mars]

Is it being too precise in Glasgow being exactly 30 degrees? Even though, yes, it is weird that it talks about Aviemore exceeding 30 degrees but treating it like Glasgow's 32 degrees means a negative answer.

That was my conclusion as well - however, it's a bit like saying Real Madrid have never scored 7 goals in a game, but they have scored 8 a few times. It's not possible to reach 31.8 deg C without it having been 30.0 deg C first.

It's really disturbing that Google is churning out obviously incorrect data as the top search result and merely leaves a 'thumbs down' option and the caveat that "AI answers might be incorrect" - I've seen multiple examples now where, given a straight binary question, it can even get those wrong too.

 

[Touring Mars]

Google will die unless it does something about it's garbage AI answers...

Asked "Why are Team Abba Racing not competing at the British GT race today?"

Google AI overview:

"British GT doesn't race on Sundays"



AI mode answer:

"They are racing today"


The only thing it has right is that AI responses may include mistakes...

 

[Imari]

Google will die unless it does something about it's garbage AI answers...

Google has been on the road to death since they decided that search volumes were more important than search quality which was way before the modern generative AI craze broke out. Bad search results like this are "good" in Google's eyes, because many users will recognise the obviously faulty information and redo their searches over and over to get the right answer and thereby expose themselves to that many more ads in doing so.

This is insane, but it also works for them because there's hardly any people willing to use a non-Google search engine even in the face of search quality that is well below what drove Google to be created in the first place. They're such a dominant monopoly that I suspect many casual users will struggle to name another search engine off the tops of their heads. It's going to take a long time before their own search quality bites them in the bum, if ever.

 

[Liquid]

Google has been on the road to death since they decided that search volumes were more important than search quality

This search of mine didn't return an AI answer, it's just straight up wrong.

 

[Danoff]

Google will die unless it does something about it's garbage AI answers...

It is making them look really bad. I ran into it just the other day as well. Again, in a medical setting:

Q: "Is Tylenol dissolve powder gluten free?"
AI: "Yes, the company has stated publicly that Tylenol dissolve powder is gluten free. Click to see more."
Q: click to see more
AI: "Yes, the company has stated publicly that Tylenol dissolve powder is gluten free. Here's a link to the source"
Q: click the source

That source? A random person leaving an Amazon review of the product.

 

[Sprite]

It is making them look really bad. I ran into it just the other day as well. Again, in a medical setting:

Q: "Is Tylenol dissolve powder gluten free?"
AI: "Yes, the company has stated publicly that Tylenol dissolve powder is gluten free. Click to see more."
Q: click to see more
AI: "Yes, the company has stated publicly that Tylenol dissolve powder is gluten free. Here's a link to the source"
Q: click the source

That source? A random person leaving an Amazon review of the product.

Won’t be long when you click for the source and it’s an official Whitehouse message.

 

[Touring Mars]

It is making them look really bad. I ran into it just the other day as well. Again, in a medical setting:

Q: "Is Tylenol dissolve powder gluten free?"
AI: "Yes, the company has stated publicly that Tylenol dissolve powder is gluten free. Click to see more."
Q: click to see more
AI: "Yes, the company has stated publicly that Tylenol dissolve powder is gluten free. Here's a link to the source"
Q: click the source

That source? A random person leaving an Amazon review of the product.

I think it will take one major law suit for Google to sort out what could be a life threateningly wrong answer.

-

What I can't stand is how it can get a straight yes or no / black or white answer just dead wrong. I get what @Imari has said above, but there's a serious moral/ethical point about making money from deliberately providing false information. That's a class action law suit in the making.

What is staggering, however, is that even with a simple 'yes or no' question, Google is still able to provide different answers that are ALL completely wrong - by getting the basic answer wrong, then providing some BS 'explanation' that is, quite literally, just a bunch of words that make sense when written in that order. It's meaningless drivel.

-

I encountered yet another example this evening, asking whether 'Gold' cards appear in booster packs in the game Balatro - yes or no. Google AI says no, they don't - plus yadayadayada ('here's why the answer is no, and it sounds convincing'); pity, because the correct answer is yes.

Won’t be long when you click for the source and it’s an official Whitehouse message.

That would figure - misinformation is the name of the game from now on.

-

@Danoff The poo emoji was, of course, not a comment on your reply - just my general feeling... and blame @Jordan for enabling those emoji sounds

 

[Danoff]

by getting the basic answer wrong, then providing some BS 'explanation' that is, quite literally, just a bunch of words that make sense when written in that order. It's meaningless drivel.

That is a huge part of it. It's not just wrong, it's so damned confident and authoritative about being wrong. That's ChatGPT as well. You ask it to generate a word file with some template info filled in, and it's like... yup, I did that, it's all filled in, here's the link to your word file.... nothing is filled in.

Don't get me wrong, when it's in the zone it's absolutely amazing. I did some astounding stuff with ChatGPT over the weekend having it generate code for me and getting it working, and it's honestly mind bogglingly good. It's just also sometimes mind bogglingly bad, and Google putting it at the top of search results is basically gross negligence.

 

[Touring Mars]

That is a huge part of it. It's not just wrong, it's so damned confident and authoritative about being wrong.

To be fair, that sounds like ALOT of people I know* 😂

* including me sometimes

That's ChatGPT as well. You ask it to generate a word file with some template info filled in, and it's like... yup, I did that, it's all filled in, here's the link to your word file.... nothing is filled in.

That bugs the hell out of me - it happens all the time. Replying "Where is the link" or "I can't download the file" generates the response "My bad, sorry - here's the file", and another ****ing dead link.

Don't get me wrong, when it's in the zone it's absolutely amazing. I did some astounding stuff with ChatGPT over the weekend having it generate code for me and getting it working, and it's honestly mind bogglingly good. It's just also sometimes mind bogglingly bad, and Google putting it at the top of search results is basically gross negligence.

Exactly - it is staggering when it gets it right, and I guess this is why it will benefit diligent users, while also destroying society at the same time. It's the intellectual equivalent of the 'rich get richer and the poor get poorer' aphorism**

** Thanks GoogleAI for that word... 😅