The two things preventing AI takeover

[Jason Voorhees]

ChatGPT builds upon existing machine learning concepts and is nothing new. These things have existed since 90s. It is only that chatgpt has optimized it, fed such a large data and made it so user friendly that even an illiterates can use it but it still suffers from Hallucination

Hallucination (artificial intelligence) - Wikipedia

en.m.wikipedia.org

And Stochastic Parroting which is a fancy way of saying the AI doesn't "understand" things the way humans do; it just predicts words based on patterns.

Stochastic parrot - Wikipedia

en.m.wikipedia.org

There have been a few proposed solutions like RLHF (Reinforcement Learning from Human Feedback) to guide Al behavior and Al guardrails fact-checking layers but they aren't foolproof and aren't reliable at all.

So we basically need a new "Einstein of Al"-like someone who can rethink the entire approach the same way Einstein flipped classical physics on it's head with relativity. Right now, Al is built on brute-force statistical learning. It mimics intelligence but doesn't understand anything. We essentially need some genius to break this deadlock and Until then, There is no AI takeover happening and you can't put a random date over when these problems will be solved. And whoever manages to solve them will go down in history as one of the greatest minds to ever exist

 

[Jason Voorhees]

@666PSL @romanstock @Sprinkles

 

[LatentIntellectual]

And Stochastic Parroting which is a fancy way of saying the AI doesn't "understand" things the way humans do; it just predicts words based on patterns.

Hence why they get the most obvious shit wrong 24/7 and will repeat the mistake again and again eveb after correction

It's also very biased towards the user. It's replies often don't try for accurate results, it just tries really hard to back up the questions asked

As it stands right now it's only a glorified smarter Google search feature with the ability to operate manual tasks

 

[Sprinkles]

All up to the kike control system

When they want something to happen, it’ll happen.

 

[Jason Voorhees]

@quantitativePSL @160cmcurry

 

[Jason Voorhees]

@Wombles

 

[quantitativePSL]

absolutely over for you if you think AI will "takeover" before you die

 

[Alias!]

@134applesauce456 do you agree with this?

 

[Bills]

Tldr this mofo thinks he's safe because we don't general artical intelligence (available to the public) yet

 

[loyolaxavvierretard]

The problem isn't with AI per say. It's that the focus is completely on LLMs and not other subjects that fall under AI. How many people investing in AI even know about the papers written in Computer Vision. Also cope if you think AI isn't the future. Nobel Prize was awarded to AlphaFold for predicting protein structures from their amino acid sequences. This was once considered impossible in biology

 

[loyolaxavvierretard]

Why is it that when you post a serious discussion nobody gives a shit but when you post some bait nonsense everyone gathers there in droves ?

 

[Jason Voorhees]

The problem isn't with AI per say. It's that the focus is completely on LLMs and not other subjects that fall under AI. How many people investing in AI even know about the papers written in Computer Vision. Also cope if you think AI isn't the future. Nobel Prize was awarded to AlphaFold for predicting protein structures from their amino acid sequences. This was once considered impossible in biology

I agree that Al isn't just about LLMs but LLMs are what dominate public perception and investment. AlphaFold is impressive but they rely on deep learning, which is still fundamentally brute-force statistical pattern matching. Even in Computer Vision most advancements are just bigger and better neural networks rather than a fundamental shift in how Al 'thinks.' LLMs as it stands are just very advanced probability machines that predict words based on past data.

 

[loyolaxavvierretard]

I agree that Al isn't just about LLMs but LLMs are what dominate public perception and investment. AlphaFold is impressive but they rely on deep learning, which is still fundamentally brute-force statistical pattern matching. Even in Computer Vision most advancements are just bigger and better neural networks rather than a fundamental shift in how Al 'thinks.' LLMs as it stands are just very advanced probability machines that predict words based on past data.

The researchers think that progressing these predictive capabilites will result in emergent properties that would mirror the properties of human brain. Let's see how right they are

 

[Muneco]

they just need to put the generational high iq jew that comes around every 30 years on this and then it's over for everyone

 

[Askeladd93270]

Why is it that when you post a serious discussion nobody gives a shit but when you post some bait nonsense everyone gathers there in droves ?

Bc when you post serious discussion some fag's report you for non seriousness so now everyone is using offtopic

 

[Jason Voorhees]

@Shahnameh

 

[Jason Voorhees]

@One Rep Max

 

[Jason Voorhees]

@moggerofhumanity @deadstock @sub5outsider @Snicket @Brus Wane

 

[borntokill]

i dont think they can take over in the near future

 

[One Rep Max]

Real problem is they’re optimised for output not cognition.
LLMs are just linguistic mirrors just following probability flows with no internal model of truth

AGI needs epistemic integrity not just coping with bigger and bigger datasets

 

[Snicket]

ChatGPT builds upon existing machine learning concepts and is nothing new. These things have existed since 90s.

My understanding is that it was the advent of the transformer in roughly 2017 by Google Engineers that revolutionized large language models and made it possible.

It is only that chatgpt has optimized it, fed such a large data and made it so user friendly that even an illiterates can use it but it still suffers from Hallucination

True. It’s not a form of intelligence, it’s basically an incredible pattern finder.

[And Stochastic Parroting which is a fancy way of saying the AI doesn't "understand" things the way humans do; it just predicts words based on patterns.

Yes. This is why Chat GPT is amazing at procedural tasks but terrible at something that requires abstract reasoning to solve.

and Until then, There is no AI takeover happening and you can't put a random date over when these problems will be solved. And whoever manages to solve them will go down in history as one of the greatest minds to ever exist

Chat GPT and other GPT based models will still put a shit load of people out work though. Successive iterations are only going to get a lot better as the training corpus increases and we GPT specialization.

By specialization, I mean you’ll end up with sector specific GPTs for finance, health etc… instead of general purpose in the next few years.

But who knows what the next decade will bring.

 

[fashioncel]

Yeah and one of them is My dreamy husband salldon

 

[Jason Voorhees]

My understanding is that it was the advent of the transformer in roughly 2017 by Google Engineers that revolutionized large language models and made it possible.


True. It’s not a form of intelligence, it’s basically an incredible pattern finder.

Yes. This is why Chat GPT is amazing at procedural tasks but terrible at something that requires abstract reasoning to solve.

Chat GPT and other GPT based models will still put a shit load of people out work though. Successive iterations are only going to get a lot better as the training corpus increases and we GPT specialization.

By specialization, I mean you’ll end up with sector specific GPTs for finance, health etc… instead of general purpose in the next few years.

But who knows what the next decade will bring.

Yeah automation is definitely going to hit a lot of jobs, especially ones that are repetitive and information-based. Sector specific GPT models will be more optimized and will perform better but even after multiple iterations the issues I mentioned will still exist. There needs to be breakthrough for AI to fully replace human decision making.

 

[Snicket]

Yeah automation is definitely going to hit a lot of jobs, especially ones that are repetitive and information-based. Sector specific GPT models will be more optimized and will perform better but even after multiple iterations the issues I mentioned will still exist. There needs to be breakthrough for AI to fully replace human decision making.

Yes, there’s still a bottleneck but I think an AI takeover, albeit limited, is a real risk and shouldn’t be understated.

 

[loyolaxavvierretard]

Yes, there’s still a bottleneck but I think an AI takeover, albeit limited, is a real risk and shouldn’t be understated.

It won't hit western countries first. They will replace all the offshore accounting and software jobs that low paid Indians do

 

[Snicket]

It won't hit western countries first. They will replace all the offshore accounting and software jobs that low paid Indians do

Hopefully. I think another real danger is a breakthrough in robotics that will hit blue collar, manual labor jobs.

 

[loyolaxavvierretard]

Hopefully. I think another real danger is a breakthrough in robotics that will hit blue collar, manual labor jobs.

That's possible but Boston Dynamics is not as open as other companies about their progress. I mean they do showcase their robots but it's feels that it's doing cherrypicked tasks.

 

[Jason Voorhees]

That's possible but Boston Dynamics is not as open as other companies about their progress. I mean they do showcase their robots but it's feels that it's doing cherrypicked tasks.

The boston dynamics dog has been running the same since 2015

 

[loyolaxavvierretard]

The boston dynamics dog has been running the same since 2015

yeah that's what i mean. They are not showing their true capabilities or they are stuck at incremental progress

 

[Snicket]

yeah that's what i mean. They are not showing their true capabilities or they are stuck at incremental progress

Language models hit a bottleneck since the late 1990s until the discovery of transformer architecture in 2017.

Breakthroughs often happen without any anticipation.

 

[loyolaxavvierretard]

Language models hit a bottleneck since the late 1990s until the breakthrough of transformer architecture in 2017.
Breakthroughs often happen without any anticipation

That's very scary. I am planning to join my father's hvac buisness. Every day I will go to sleep dreading what technology they are cooking in their labs to replace blue collar folks like me

 

[Snicket]

That's very scary. I am planning to join my father's hvac buisness. Every day I will go to sleep dreading what technology they are cooking in their labs to replace blue collar folks like me

Defaulting to the family business is always a safer option. Wish I had it.

 

[loyolaxavvierretard]

Defaulting to the family business is always a safer option. Wish I had it.

Yeah it's nice. But nothing is guaranteed nowadays in this field, especially with bigger players entering the market

 

[134applesauce456]

my iq isn't high enough for this discussion but bump

 

[Shahnameh]

Chatgpt is a glorified search engine thats reactive based on user generated prompts like any other LLM, even to run this thr server costs are astronomical and as the LLM gets more and more advanced so will the costs.

Its not an Ai self thinking machine, boomers and idiots hype it up and think its going to replace all jobs which is a load of shit.

Ai is just a very useful buzzword for investors, it's basically the new Crypto and NFT

 

[Offensive Bias]

They would likely need to utilize some wetware based system in conjunction with existing LLMs to overcome the cognition hurdle.