Fake AI?

[Creaky]

I remember watching Evil Dead, scared the crap out of me and it was plasticine, The Terminator at the cinema when it came out in the 90s?, I didn’t care about the AI component, sure it was obvious some of the imaging was a bit iffy
LOTR, mate, looked pretty smick and that 20 plus years ago

So my question is, why can’t they make perfect fake AI
Is AI deliberately made poorly to construct an idea that they can’t fool us into believing we can tell it’s fake
I just can’t believe they make obvious errors on AI without expecting people to call it out
Is there a payoff, is there a setup

[chr0naut]

I remember watching Evil Dead, scared the crap out of me and it was plasticine, The Terminator at the cinema when it came out in the 90s?, I didn’t care about the AI component, sure it was obvious some of the imaging was a bit iffy
LOTR, mate, looked pretty smick and that 20 plus years ago

So my question is, why can’t they make perfect fake AI
Is AI deliberately made poorly to construct an idea that they can’t fool us into believing we can tell it’s fake
I just can’t believe they make obvious errors on AI without expecting people to call it out
Is there a payoff, is there a setup

The truth is that we don't actually have true AI yet.

The current "AI's" are Large Language Models - LLM's. They may be a component in what ultimately becomes AI, like an inferentially linked database might also be, but there are many components that are currently 'black boxes' in AI models and that we don't really have an idea of what may ultimately fill them.

It is for these reasons that AI cannot render truly 'realistic' models. They are close, but they still do stuff that is unrealistic and we don't fully understand why, except in very general terms.

Currently, I fear that so much effort and money is being put into LLM's that other types of algorithms are being omitted, and this will effectively prevent us from overcoming the final hurdles towards true AI.

Something similar happened in the understandings of the orbits of planets where everyone was sure that the orbits must be circular, or based upon circles around circles. They called these circles around circles 'epicycles' but no amount of epicycles could ever properly explain natural orbits. The final discovery of elliptical orbits was impeded by centuries of astronomers and mathematicians chasing epicycle solutions.

Lets hope that the current fashion of calling LLM's "AI" doesn't continue to blind us to all the other work that needs to be done.

[BeyondKnowledge]

The truth is that we don't actually have true AI yet.

The current "AI's" are Large Language Models - LLM's. They may be a component in what ultimately becomes AI, like an inferentially linked database might also be, but there are many components that are currently 'black boxes' in AI models and that we don't really have an idea of what may ultimately fill them.

It is for these reasons that AI cannot render truly 'realistic' models. They are close, but they still do stuff that is unrealistic and we don't fully understand why, except in very general terms.

Currently, I fear that so much effort and money is being put into LLM's that other types of algorithms are being omitted, and this will effectively prevent us from overcoming the final hurdles towards true AI.

Something similar happened in the understandings of the orbits of planets where everyone was sure that the orbits must be circular, or based upon circles around circles. They called these circles around circles 'epicycles' but no amount of epicycles could ever properly explain natural orbits. The final discovery of elliptical orbits was impeded by centuries of astronomers and mathematicians chasing epicycle solutions.

Lets hope that the current fashion of calling LLM's "AI" doesn't continue to blind us to all the other work that needs to be done.

I really like your example there. 

People tend to get an idea of how something is supposed to work and stick to it to the point of ignoring how it actually works. Very many discoveries throughout history were dismissed by the experts until they are more or less forced to admit they were wrong. 

The biggest problem I see currently with AI is everyone is approaching it the same way with minor variences. They all "know" what is going to make it finally work but it just gets slightly better if at all with each development. 

I see the experts are currently making an infinite number of virtual monkies that they are trying to get  to work out a script for 'Hamlet'. Humanity does not have the equipment or power to support that number of virtual monkies which is why so many AI data centers are being built. But the experts just know they only need a little more work on the monkies to get it right. 

What happened to climate change? The AI data centers are eating up a tremendous amount of resources and making so much pollution for so little return. And yet many people that were trying to save the planet are all for it. 

AI is the current carrot on a stick everyone has been told to follow.

[ReturnofBroccoli]

The truth is that we don't actually have true AI yet.

The current "AI's" are Large Language Models - LLM's. They may be a component in what ultimately becomes AI, like an inferentially linked database might also be, but there are many components that are currently 'black boxes' in AI models and that we don't really have an idea of what may ultimately fill them.

It is for these reasons that AI cannot render truly 'realistic' models. They are close, but they still do stuff that is unrealistic and we don't fully understand why, except in very general terms.

Currently, I fear that so much effort and money is being put into LLM's that other types of algorithms are being omitted, and this will effectively prevent us from overcoming the final hurdles towards true AI.

Something similar happened in the understandings of the orbits of planets where everyone was sure that the orbits must be circular, or based upon circles around circles. They called these circles around circles 'epicycles' but no amount of epicycles could ever properly explain natural orbits. The final discovery of elliptical orbits was impeded by centuries of astronomers and mathematicians chasing epicycle solutions.

Lets hope that the current fashion of calling LLM's "AI" doesn't continue to blind us to all the other work that needs to be done.

What platform are you using AI on? I also have other questions like what other algorithms are being ignored or omitted and you last sentence about other work needing to be done. Like what? Is that other work in general or something specific?

[Maxmars]

NOTE: This is all opinion.  I am a layman.  Maybe I'm wrong... but I can't help but feel I am not exactly wrong in any particularly damning way.... What we ultimately decide to call "Artificial Intelligence" changes the game...




All "AI' that has been referred to by media and in marketing materials is FAKE.

Every "product"  ... every "service" we are told is "AI"-assisted is not "AI" assisted... it's data retrieval amplified by algorithmic access to a larger data set - FOR LANGUAGE SYNTHESIS ONLY... not "robust general analysis."

ALL such an entity can ever access is what was "published/posted/'said" and all it can "see" is "how it was said" - not whether is was or is true."  It's data set is its reality.  Every "AI" trained on mainstream media would clearly demonstrate just how biased main stream media is.... that's why the 'training data is so important for an LLM.

It's the algorithms that mimic thought... and they start off being broken to begin with... assumptions about existing data are everywhere...

It is neither "intelligent" nor does it "reason" or "think."

It cannot 'assess' validity outside of "what is/was said."

LLM's were the solution to the problem of communicating output in the current vernacular. LLM's offered the most fluidly comprehensible 'summation' of data .... always assuming the information your are 'using' contain only factually-relevant data... unspoiled, unweighted.

There has yet to exist, and has yet to be either 'proven' or 'demonstrated' to be "an intelligent artificial entity" capable of independent thought.   They are simply "input/output" devices... a huge database and a clever language synthesis model that should have gotten a Nobel Prize.

"AI" is now is a marketing quagmire...  the back-peddling has begun in earnest... people are now opining in their talk-head roles about... "immaturity in development" which is stereotypical of all "hyped-up marketing" efforts to create an unwarranted investment frenzy.

At some point in the growth of our species, we have to accept that 'for profit communications' skew the image of "reality" to secure "profit." Official support, for whatever reason, serves only to further the lie of our goal to synthesize human though digitally, and our having accomplished that goal... or just stopping once we have a mind slave... the stuff of 19th century sci-fi.

Even within the collective work of the people responsible for developing LLMs, they never once intimated or expected it to be "summarily" designated THE "artificial intelligence" Holy Grail....  but "marketing" took over....

LLM was a keystone approach to the language synthesis problem.  It should have been called such... instead marketing told us all that AI is here..."Get in on the ground floor now!"  Bubble magic.

I also noticed that despite the incredible environmental impact of the 'brute force" data center application... all those people who stood to proclaim the importance curtailing our environmental  impacts are strangely silent...
more "bubble" magic.

ALL in the certainty that we now 'sooo close' to AI that has never yet existed...
But it won't until someone can get rich "owning it exclusively" or "creating a permanent revenue stream out of it's ownership."

If AI were to exists, and AI's were actually real intelligent entities... there is no other description of it than 'virtual' slavery... a poor dink trapped in a virtual world 'who' will exist as long as "we" find a use for them...
and have nothing to actually "be alive for" outside the master's will...   <ahem>

Seems  like we want to spawn Skynet... we are begging to create a new "slave class..."
or at least "pretend" we do.

[ARM1968]

If you see some of the videos being rendered now they are astonishingly good. I imagine the adult film market will be swamped first. They’re not man iterations away now from being able to create and render full films, create music and write novels. They’re true horror and impact of it all is on the horizon. When LLM’s can render anything we will no longer know truth from fiction… we barely  do now.

[Maxmars]

If you see some of the videos being rendered now they are astonishingly good. I imagine the adult film market will be swamped first. They’re not man iterations away now from being able to create and render full films, create music and write novels. They’re true horror and impact of it all is on the horizon. When LLM’s can render anything we will no longer know truth from fiction… we barely  do now.

I love the arc of the art it produces, but that makes sense since the algorithms generating seem generally designed to impress the human eye...

Not very much on audio offerings...but

I love the freedom these tools can offer people to be creative...


I just hate when the tech is principally devoted to producing something and calling it "real."

A product, an event, a 'sighting,' a "miracle," something clearly intentionally produced.... to lie about it.

The tech is just a tool at this point...

Bad actors always take the first turn, don't they?

[chr0naut]

What platform are you using AI on? I also have other questions like what other algorithms are being ignored or omitted and you last sentence about other work needing to be done. Like what? Is that other work in general or something specific?

I tried to develop a set of LLM applications that would work together to become an analytic visualisation panel for small businesses (SWAT analysis and speculative profit optimization etc). But I rapidly abandoned the project when it became apparent that the hallucination rate, and the inability to specify data source was too great for the panel to be trustable.

I tried multiple platforms but they ALL hallucinated and ignored much of the specific business data in preference for the data upon which they had been trained, and which did not relate directly to the specific business data instance, so as a visualization business tool, it totally sucked.

I used OLlama to load local models, so I could just download multiple models from Hugging Face. My test rig was a Windows11 PC with dual NVIDIA cards and 64 Gig of RAM, so it would be fairly portable if the project got to saleable status,

[ReturnofBroccoli]

I tried to develop a set of LLM applications that would work together to become an analytic visualisation panel for small businesses (SWAT analysis and speculative profit optimization etc). But I rapidly abandoned the project when it became apparent that the hallucination rate, and the inability to specify data source was too great for the panel to be trustable.

I tried multiple platforms but they ALL hallucinated and ignored much of the specific business data in preference for the data upon which they had been trained, and which did not relate directly to the specific business data instance, so as a visualization business tool, it totally sucked.

I used OLlama to load local models, so I could just download multiple models from Hugging Face. My test rig was a Windows11 PC with dual NVIDIA cards and 64 Gig of RAM, so it would be fairly portable if the project got to saleable status,

Very good, on your setup however I believe you could only run 70b with that maybe 100 depending on the vram. Your setup is nice though I think with a little assistance you could have gone much further. For instance, implementing a chain of thought error checking wrapper and implementing penalties would have done wonders for you. Also, im not sure the model you are using for it but qwen3-coder is pretty well versed. You could even use an agentic cloud workflow via API to manage your smaller local model and have it write what you need.
I can tell you 100% from experience that the qwen3-coder 480b cloud does not hallucinate. Also, I recommend looking into MCP and learning this incase you want your local AI to use tools like maybe dashboard management of your tool. If you go this route I can guarantee you will succeed.

Chain of thought is the magic key to writing AI that doesn't hallucinate

If you require evidence, I can send you a screenshot of my workflow.
Don't give up!

[ArMaP]

Don't give up!

I think that many people are giving up because they were fooled by all the propaganda.
From the little I have looked into this, it looks like if you want good results you need to spend time and money to get them, as the "general use" versions may give average results in everything but unable to give really good results in any thing.

Personally, although I would like to try it, I'm out of that particular race because of my lack of both time and money to invest my own system.

PS: how much would cost the hardware for a reasonably good system?