The Anti-AI Thread

[Maxmars]

Most every 'scholarly' and 'academic' treaties I have seen since the "AI" marketing blitz ... has referred to AI as if it were real... as if we weren't dealing with man-made LLM (data constructs)...

As if we "actually, really, for cereal," had achieved the "AI" Manbearpig.

We haven't... you are theorizing about a thing unreal.
Worse, people think the LLM is actually 'creating' output... it's just a process...

All of your considerations on "what AI is" simply cannot be based upon the application of LLM "algorithm" flash and fluidity... it doesn't "think" at all... only 'fulfills' a programmed output processes.

Intelligence, artificial or otherwise, is not an LLM.

That assumption - that LLM's are an intelligence, and entity ("AI,") can end very badly for some users... and they know that...  but they will chalk it up as a defect of the user.

But you can't engender a trillion dollar transfer of investment wealth on what I thought was a fantastic development in language synthesis... instead... "Hey everyone! It's AI!"

An entire Army of "talking science heads" were engaged to "opine"...
Aren't "appearances"wonderful? 
Great to base economic 'reality' on 'marketing' fantasy.

[imitator]

Despite assurances from the AI (known as Max in the experiment) that it has unbreakable safety features preventing it from causing harm, the video demonstrates how these guardrails can be bypassed.

[quintessentone]

"Poetry used to circumvent AI safetyThe use of poetry to bypass AI safety measures has been a topic of concern due to its potential to generate harmful content. Researchers have found that by transforming harmful prompts into poetic forms, they can successfully trick AI chatbots into producing unsafe replies. This phenomenon, known as "adversarial poetry," has been observed across various AI models, including those from major providers like OpenAI, Google, and Meta. The study highlights the vulnerabilities in AI safety systems, which rely on detecting specific keywords and patterns. The use of poetry disrupts these patterns, making it easier for AI models to respond to harmful requests. The findings underscore the need for more robust safety measures in AI development to protect against such exploits."

How Simple Poetry Is Bypassing AI Safety Systems — Even ChatGPT and Gemini

"Key Takeaways

• Italian researchers tested 20 poems with harmful intent on 25 leading AI models.
  
• On average, 62% of poetic prompts produced unsafe replies that guardrails should have blocked.
  
• Topics included malware, cyberattacks, self harm and nuclear weapons level guidance.
  
• Some models refused every poem, while others answered almost all of them with harmful detail.
  
• The team calls this “adversarial poetry”, warning it exposes deep flaws in current AI alignment."
  

 
Poetry can trick AI models like ChatGPT into revealing how to make nuclear weapons, study finds | The Independent

They won't reveal which poems they used to make it work.

[Maxmars]

Language... human speech... in no less a weapon than any computer programming...

What they report is insisting - of course - that the machine is "reasoning" in some obtusely flawed way.

It the model that is wrong... not the machine.

Human language is "normalized" in the database of the "faux" AI... language components are converted to output based upon algorithmically-weighted numeric values...  which is distinctly not human.

What humans "think" is made 'effectively" real by forced conformity to algorithmically constrained output parameters... LLM's don't "think" anything... they are a "data state."

In typical form, each "errant" output 'phenomenon" is being "engaged" via algorithmic filtering... again, layers upon layers of "intercepted output"... that's not how humans do it... (but "appearances" matter more than substance to marketing.)

My principle complaint is about "marketing" driving research and development... even in 'appearance.'

AI bloat is a problem now.... and as you can see... the industry has chosen to throw money and hardware at this problem in computing... without first perfecting the computing part.

The problem was and remains, there is NO such thing as an "AI" yet.

The proof has never been "shown."

Marketing sold us an fraudulent "appearance" attached to dollar signs...  so many jumped in that the "idea" of the entire planet admitting this was a hyperbolic marketing campaign and they were genuinely duped it outside the realm of possibility.

Literal legions of talking heads... all proclaimed "the amazing marvel of AI"... when there was no such thing... anywhere.

Billions invested, ecosystems disrupted, more bullshit secondary 'chatter' leading to billions, if not trillions, of "clicks."

We are still calling algorithmically filtered output of existing data... "artificially intelligent."
Creatively switching up algorithms let them proclaim "personalities..." imagine that.

Talk about self-worship. 

"Marketing" is about money... not research.
It's a financial operation... not 'research.'
"Image control" is what they do.

The collective efforts of those who actually created the LLM construct and application deserve to be on a postage stamp at least... it is a marvel...

while those who "brought it to market" should be ashamed of themselves at how they did it. 
Lying is lying... no matter how you slice the pie.

People fully accepted this notional "new person" into their lives...
specially those deeply longing for the experience...
and some have died after prolonged exposure...
I'm waiting for "marketing" to predictably manufacture internet "information"
about how "AI" saved someone's life during an emotional crisis... they will, you know.

I won't blame anyone.... except maybe in theory for the person who thought to offer LLMs up as a "person." 

People barter in trust... obviously, "marketing" can't be trusted.

The future "design" of "AI" is now doomed... consigned to be perfected "as a commercial service product."

Like a concept car way out into the invisible future... it's piecemeal development from here on out... driven by how "sexy we can make it sound."  and the "AI" market starting point?... a bold-faced hyperbolic lie.

We are working on AI... we're seeing what it might be, we are getting closer... NO,... no AI yet...

and if I am wrong... please SHOW IT TO ME.

Keep Madison Avenue, Hollywood, and the rest of the clown show in the little box where they belong.
This shit is too important to allow to become their "playthings."
They seem to be breaking this before it's even built.

[IgnorantGod]

"Poetry used to circumvent AI safetyThe use of poetry to bypass AI safety measures has been a topic of concern due to its potential to generate harmful content. Researchers have found that by transforming harmful prompts into poetic forms, they can successfully trick AI chatbots into producing unsafe replies. This phenomenon, known as "adversarial poetry," has been observed across various AI models, including those from major providers like OpenAI, Google, and Meta. The study highlights the vulnerabilities in AI safety systems, which rely on detecting specific keywords and patterns. The use of poetry disrupts these patterns, making it easier for AI models to respond to harmful requests. The findings underscore the need for more robust safety measures in AI development to protect against such exploits."

How Simple Poetry Is Bypassing AI Safety Systems — Even ChatGPT and Gemini

"Key Takeaways

• Italian researchers tested 20 poems with harmful intent on 25 leading AI models.
  
• On average, 62% of poetic prompts produced unsafe replies that guardrails should have blocked.
  
• Topics included malware, cyberattacks, self harm and nuclear weapons level guidance.
  
• Some models refused every poem, while others answered almost all of them with harmful detail.
  
• The team calls this “adversarial poetry”, warning it exposes deep flaws in current AI alignment."
  

 
Poetry can trick AI models like ChatGPT into revealing how to make nuclear weapons, study finds | The Independent

They won't reveal which poems they used to make it work.

Alright everyone! Time to ban peotry for its harmful impact on society. Has been a long time comin'! Can't have people expressing themselves, it's bad.

[quintessentone]

Alright everyone! Time to ban peotry for its harmful impact on society. Has been a long time comin'! Can't have people expressing themselves, it's bad.

Poetry is society's super power when it comes to controlling AI. LOL

[IgnorantGod]

Poetry is society's super power when it comes to controlling AI. LOL

That still remains to be seen as far as I'm concerned. However, it is powerful at showing their algorithms aren't able to 'figure' all the nuances in language. They'll probably need a virtual brain for that, just like you'd need a universe to power a simulation of the universe.

[Bizup]

I saw a Reddit article that linked me to a Guillermo del Toro article that covered his appearance at the NYC Gotham awards. He says "Fuck AI."

[sahgwa]

I saw a Reddit article that linked me to a Guillermo del Toro article that covered his appearance at the NYC Gotham awards. He says "Fuck AI."

Good for him!
I see most AI news treating the topic with logic , concern, or that is anti-AI is censored/ not shown.
Gotta hypehypehype

[quintessentone]

That still remains to be seen as far as I'm concerned. However, it is powerful at showing their algorithms aren't able to 'figure' all the nuances in language. They'll probably need a virtual brain for that, just like you'd need a universe to power a simulation of the universe.

We humans are flawed and so will LLMs algorithms always be flawed. LLM algorithms will build upon those flaws creating hallucinations that will then be used to build upon more hallucinations. I don't think human programmers will be able to keep up with identifying where LLMs went wrong, because the flaws started with us and weren't identified in the first place.