The A.I. Thread

[Fuzz]

It's not weird or flawed. It's a narrow study. To me it looks well designed, and they even made sure the solution was one of the first Google results. It's showing they are worse than Google at that specific type of search. You don't get to call it a ####ty study because they made specific choices to test one thing. Do you understand how science works?

And FWIW there are many times I take a snippet of text and search for it to find the source. This isn't an unusual thing to do. And since you want a link to the primary source, it's valuable to get it right. This study shows you can't trust an LLM to do that reliably.

Also, your stock search came back with flawed results. I'd be interested to know where that (4 or 6) link leads to, as Tesla closed at 222, not 227. And if you have to fact check all the sources, is it really providing value?

[Firebot]

Originally Posted by Fuzz

Also, your stock search came back with flawed results. I'd be interested to know where that (4 or 6) link leads to, as Tesla closed at 222, not 227. And if you have to fact check all the sources, is it really providing value?

Good catch and I stand corrected. I used the regular Perplexity (auto) for the search I posted but it is indeed wrong.

https://www.investopedia.com/tesla-s...onday-11693811

The stock finished down more than 15% to just over $222, its lowest level since late last year, pulling the company's market capitalization to around $840 billion, according to Visible Alpha data. Equity markets were down broadly in today's trading.

This is where it pulled it from or at least claiming to cite from (which obviously it doesn't exist). Ironically, 2 other attempts got the right 222.15$, and the one I did on Perplexity Pro prior to my post had it right as well. Still, if it's not accurate every time it defeats it's use. It's possible it may also get results wrong with Perplexity Pro.

"All Bad at Citing News" which the study concluded is accurate, so for what it was looking at (narrow use that is likely to fail) it did good.

[Fuzz]

I only remembered it was $222 because I was waiting for it to bust throguh that 220 "support"

[Fuzz]

I had posted about this a couple weeks ago, now there is some research into Russia poisoning AI data sources.

The network—dubbed "Pravda," a nod to the Soviet-era newspaper—has been systematically injecting AI chatbots with false narratives by gaming search engines and web crawlers. The implications are severe: AI models are increasingly echoing Kremlin-backed falsehoods, compromising information integrity at an unprecedented scale.

NewsGuard’s audit of 10 leading generative AI tools—including OpenAI’s ChatGPT-4o, Google’s Gemini, and Microsoft’s Copilot—found that the models repeated Pravda’s false narratives 33 percent of the time.

John Mark Dougan, an American fugitive turned Kremlin propagandist, laid out this strategy bluntly in a Moscow conference earlier this year: “By pushing these Russian narratives from the Russian perspective, we can actually change worldwide AI.” Dougan’s statement underscores a key objective of the Pravda network: weaponizing AI-generated content to reshape global narratives in Russia’s favor.

It syndicates misleading content across 150 seemingly independent websites, each optimized for AI and search engine algorithms. This includes fabricated claims about Ukrainian President Volodymyr Zelensky misappropriating military aid and false reports of U.S. bioweapons labs in Ukraine.

NewsGuard’s findings align with those of the American Sunlight Project (ASP), a U.S. nonprofit that has termed this strategy “LLM grooming.” According to ASP, the more frequently a false narrative appears in search results and indexed content, the greater the likelihood that large language models (LLMs) will absorb and regurgitate it.

“The long-term risks—political, social, and technological—associated with potential LLM grooming within this network are high,” the ASP report warned. “The larger a set of pro-Russia narratives is, the more likely it is to be integrated into an LLM.”

https://www.enterprisesecuritytech.c...tern-ai-models

Russia is winning a war we didn't even know was happening.

[Russic]

The issue we're seeing here (to me) really outlines a reason a lot of us are going to be blindsided by all this. It's really good at some things, and really bad at others. If what you do overlaps with the latter, you're likely to dismiss it and stop paying attention (which is reasonable).

For example, what Firebot and I do almost never require that needle-in-a-haystack search capability, but the same can't be said of a lawyer (I presume), who might need damn accurate search capability.

The advancement moves so quick though, and a ton of early LLM issues have been fixed for over a year and nobody even noticed (it's very common to hear people say they can't do math when they're well into solving PhD level math problems by this point).

I assume history will repeat and this search issue will be lessened over time.

[Fuzz]

Ya, there are great use cases. I get nervous when companies and governments decide to use them because they have only listened to the good bits. I thinks it's worthwhile to know where they are not good to avoid making big, or just embarrassing mistakes by using their output unchecked.


It's like a Tesla self driving car. It's pretty great and reliable most of the time. Enough you can trust it to make the right decisions...until it kills you one day because you trusted it. LLM's probably aren't going to kill you, but being aware of their limitations is important. And for the developers of them, knowing what needs work is the valuable bit.


Ultimately I think they are flawed for the same reasons humans are. Overconfidence and reliance on knowledge. If the information is wrong, they can never reason out the correct answer. And as we well know, you can't trust everything you read on the internet. The overconfidence can be dialed down, but verifying facts on which decisions are made is a lot harder nut to crack.

[Firebot]

A little update on my own personal projects using solely AI, I am now right around at the ~5000$ USD mark since I started this escapade mainly from one stream. But ironically, this isn't the big one at all anymore.

I recently became viral on TikTok on a new account, getting 50K+ followers since January and over 30 million views in that span. My videos are consistently at the top of high traffic search terms and completely took over the search and fyp as the top Tik Tok creator for those words. This thing came out of the blue to be honest but I analyzed what was consistently popular and made a better version of it, effectively creating a new genre within it and it took off.

I'm basically scrambling to have my AI business website fully fledged as I'm getting sponsored video requests from some of the bigger AI video company names and recently accepted in Tik Tok One. I also got recruited and accepted as an Amazon Influencer which will help me expand and keep monetizing should Tik Tok get shut off. All of a sudden I now have an audience to promote my own apps. My efforts are going here right now to monetize as quickly as I can to have a cross platform presence.

My videos are generally created with a combination of Perplexity (Deepseek R1) + Midjourney + Kling + upscaler Topaz AI / video editing Davinci Resolve, Suno for music and ElevenLabs for voiceover.

Stay tuned.

[Russic]

Originally Posted by Fuzz

Ya, there are great use cases. I get nervous when companies and governments decide to use them because they have only listened to the good bits. I thinks it's worthwhile to know where they are not good to avoid making big, or just embarrassing mistakes by using their output unchecked.


It's like a Tesla self driving car. It's pretty great and reliable most of the time. Enough you can trust it to make the right decisions...until it kills you one day because you trusted it. LLM's probably aren't going to kill you, but being aware of their limitations is important. And for the developers of them, knowing what needs work is the valuable bit.


Ultimately I think they are flawed for the same reasons humans are. Overconfidence and reliance on knowledge. If the information is wrong, they can never reason out the correct answer. And as we well know, you can't trust everything you read on the internet. The overconfidence can be dialed down, but verifying facts on which decisions are made is a lot harder nut to crack.

To me you've hit the nail on the head. It gets scary when these things make weird stupid error, and then we want to make them do air traffic control.

But we run into a strange conundrum when we compare them to humans, because we're actually quite terrible at many things. My doctor friend has quietly confided in me that her patient diagnoses are rapidly improving, and it illuminated her to "2 very serious issues I didn't have on my radar" for a couple patients.

The human psychology angle of AI potentially saving 1000 lives while simultaneously doing something dumb and killing 100 people is fascinating to me.

Originally Posted by Firebot

A little update on my own personal projects using solely AI, I am now right around at the ~5000$ USD mark since I started this escapade mainly from one stream. But ironically, this isn't the big one at all anymore.

I recently became viral on TikTok on a new account, getting 50K+ followers since January and over 30 million views in that span. My videos are consistently at the top of high traffic search terms and completely took over the search and fyp as the top Tik Tok creator for those words. This thing came out of the blue to be honest but I analyzed what was consistently popular and made a better version of it, effectively creating a new genre within it and it took off.

I'm basically scrambling to have my AI business website fully fledged as I'm getting sponsored video requests from some of the bigger AI video company names and recently accepted in Tik Tok One. I also got recruited and accepted as an Amazon Influencer which will help me expand and keep monetizing should Tik Tok get shut off. All of a sudden I now have an audience to promote my own apps. My efforts are going here right now to monetize as quickly as I can to have a cross platform presence.

My videos are generally created with a combination of Perplexity (Deepseek R1) + Midjourney + Kling + upscaler Topaz AI / video editing Davinci Resolve, Suno for music and ElevenLabs for voiceover.

Stay tuned.

Lmao, from such humble beginnings! For all the debate online about "is it a bubble", there are simply people who merely go out and do stuff. I think those with agency and curiosity stand to make some very interesting things.

Speaking as somebody in the entrepreneurial space (kinda), an uncomfortable truth is you need to take a lot of swings at different stuff then iterate on what works over and over again. It can be a massive investment in many ways.

This though? Somebody like me (and I mean that in the most derogatory, self-loathing way) can spin up probably 10 concepts in a month. I don't think people appreciate that (as Firebot has shown), finding a successful thing is often a factor of just swinging until luck finds you.

[Fuzz]

Ya, like a self driving car is gongs to have to be safer than a human for us to accept them. I think the difference between an individual human making a mistake and AI is that a human is one, and the AI system is multiple. So if you took 1000 human drivers they'd almost all have a few things they just suck at, but those things would be different, for the most part. And you may have a few that are near perfect in their actions/choices. It's easier to accept one human making one mistake than it is to accept an AI making the same mistake over and over.



So I think it necessarily needs to be held to a higher standard, to that near perfect human, or an aggregate of all the good decisions those 1000 humans make. Essentially it's like AI can commit the mistake one human makes once, and we would be OK with that. But if it makes the same mistakes as 1000 humans, while also being as good or better on the positive side, it's still a complete failure. Which I think is fair, because statistically you will interact with a different fialing human rarely, but if the AI fails you will be interacting with it regularly, given it's reach.

[GGG]

Originally Posted by Fuzz

Ya, like a self driving car is gongs to have to be safer than a human for us to accept them. I think the difference between an individual human making a mistake and AI is that a human is one, and the AI system is multiple. So if you took 1000 human drivers they'd almost all have a few things they just suck at, but those things would be different, for the most part. And you may have a few that are near perfect in their actions/choices. It's easier to accept one human making one mistake than it is to accept an AI making the same mistake over and over.



So I think it necessarily needs to be held to a higher standard, to that near perfect human, or an aggregate of all the good decisions those 1000 humans make. Essentially it's like AI can commit the mistake one human makes once, and we would be OK with that. But if it makes the same mistakes as 1000 humans, while also being as good or better on the positive side, it's still a complete failure. Which I think is fair, because statistically you will interact with a different fialing human rarely, but if the AI fails you will be interacting with it regularly, given it's reach.

In the driving space its errors that a human wouldn’t have made that I think people find troubling. People only see the deaths it causes rather than the ones it prevents. I think it’s still just an insurance problem. If insurance companies insure cars in the same manner we do now eventuallly costs will be lower than humans and it will win despite the objections.

In AI I think you bring up a really good point. Our society is designed for random error it is not designed for repetitive errors. As errors become less random and more systematic all of the current quality systems will break down. A random sampling of 5% of work no longer catches errors. It’s now needs 100% checks until you prove a process is good and then 0. It’s a completely different quality paradigm.

[jayswin]

More importantly than any and all of this is that we needed to figure out how to support societies as AI takes over jobs.......And we're now realizing we have no plans and billion dollar businesses are not going to do **** as early optimistic projections of AI takeover assumed.

Countries will be in no position to so so either. So where are we? We're ****ed.

[jayswin]

I appreciate Firebot and Russic's passionate enthusiasm for AI as they're in the industry and having fun and deep in the excitement of what you can do with it, but I wouldn't be following their excitement if I was....most human beings working.

[Russic]

Originally Posted by jayswin

More importantly than any and all of this is that we needed to figure out how to support societies as AI takes over jobs.......And we're now realizing we have no plans and billion dollar businesses are not going to do **** as early optimistic projections of AI takeover assumed.

Countries will be in no position to so so either. So where are we? We're ****ed.

Originally Posted by jayswin

I appreciate Firebot and Russic's passionate enthusiasm for AI as they're in the industry and having fun and deep in the excitement of what you can do with it, but I wouldn't be following their excitement if I was....most human beings working.

Ya I see nothing to disagree with here. Government already moves at a snail's pace, and all signs are pointing to a very rapid change that will require adjustment. There are more than a few canary's in the coal mine.

Computer Science jobs dropping off a cliff is definitely one of them, and we can watch those job postings dry up in real time. I should say, I only know about this from people posting online, so if I'm wrong, please fact-check me. As far as I know, I can't name an AI tool that can replace a half-competent developer yet. Perhaps by the end of 2025 with the agents we're starting to see, but certainly not over the last year.

It's also important to note, I'm generally optimistic by nature. I also don't consider myself very good at forecasting anything. The number of times I was certain of something becoming a massive problem or outrageous success only to be proven super wrong is incredible.

For that reason, while it does indeed seem hopeless at times, there's thousands upon thousands of micro developments that will change the trajectory of things in the future. I choose to be hopeful because frankly it's far more enjoyable and history tends to be on its side.

I will also absolutely own up to the fact that my job is on the chopping block and every now and then that dread certainly creeps in.

[Itse]

Originally Posted by Fuzz

Ya, there are great use cases. I get nervous when companies and governments decide to use them because they have only listened to the good bits. I thinks it's worthwhile to know where they are not good to avoid making big, or just embarrassing mistakes by using their output unchecked.


It's like a Tesla self driving car. It's pretty great and reliable most of the time. Enough you can trust it to make the right decisions...until it kills you one day because you trusted it.

Considering they cause an estimated 10 times more accidents than US human drivers, who are already incredibly accident prone compared to European drivers... Yeah.

Elon Musk should absolutely be in jail just for the deaths he caused by pushing unsafe technology into traffic and blatantly just lying about it's capabilities all the time.

[Bill Bumface]

Originally Posted by Russic

Computer Science jobs dropping off a cliff is definitely one of them, and we can watch those job postings dry up in real time.

I'm not as worried. There was a massive industry bubble that the pandemic pushed to its bursting/deflating point that is going on regardless of AI.

For some early stage prototyping stuff, sure some jobs may be drying up, but that whole sector is dry due to lack of funding now anyway.

We're not seeing the massive productivity increases that are shrinking teams dramatically, but we are seeing nervousness in hiring in case that happens, which is very different at this point.

Industrial agriculture killed lots of farming jobs, and other sectors emerged to employ those people. Computers were supposed to eliminate tons of jobs. Now we just have a massive amount of people working on computers and the tech that runs on them instead.

[Russic]

Originally Posted by Bill Bumface

I'm not as worried. There was a massive industry bubble that the pandemic pushed to its bursting/deflating point that is going on regardless of AI.

For some early stage prototyping stuff, sure some jobs may be drying up, but that whole sector is dry due to lack of funding now anyway.

We're not seeing the massive productivity increases that are shrinking teams dramatically, but we are seeing nervousness in hiring in case that happens, which is very different at this point.

Industrial agriculture killed lots of farming jobs, and other sectors emerged to employ those people. Computers were supposed to eliminate tons of jobs. Now we just have a massive amount of people working on computers and the tech that runs on them instead.

This is what it feels like to me somewhat. Zuckerberg running around saying there won't be a need for developers in a year (or whatever his quote was) is definitely not helping, especially when it's people like him that stand to gain from hyping these things.

Like I said, I haven't seen anything to indicate we can replace a developer with AI yet. I've seen a developer 3x his output, but then he seems to be the only one at his work actively using it. The other developers in my circle seem to either hate it or pay it no mind.

[Fuzz]

What has AI been used for this week?

People Are Using AI to Create Influencers With Down Syndrome Who Sell Nudes

https://www.404media.co/people-are-u...ho-sell-nudes/


OK, that just about does it for humanity.

[Fuzz]

Majority of AI Researchers Say Tech Industry Is Pouring Billions Into a Dead End
"The vast investments in scaling... always seemed to me to be misplaced."

This, more or less, is the line being taken by AI researchers in a recent survey. Asked whether "scaling up" current AI approaches could lead to achieving artificial general intelligence (AGI), or a general purpose AI that matches or surpasses human cognition, an overwhelming 76 percent of respondents said it was "unlikely" or "very unlikely" to succeed.

https://futurism.com/ai-researchers-...ustry-dead-end


I wonder how much closer we are to AGI than we were 10 years ago. Has way too much money been chasing a dead end? I'm not at all suggesting current models aren't useful, but I've never really bought into the idea that they were the path to AGI any more than the Mechanical Turk was. They are just getting to be better imitators.


I guess the reality is, perhaps given the current global situation, that it's better AGI doesn't exist. We are well on our way to a dystopian catastrophe, adding AGI into it when the US government is a puppet of the technocracy seems a very dangerous game.

[GGG]

Originally Posted by Fuzz

https://futurism.com/ai-researchers-...ustry-dead-end


I wonder how much closer we are to AGI than we were 10 years ago. Has way too much money been chasing a dead end? I'm not at all suggesting current models aren't useful, but I've never really bought into the idea that they were the path to AGI any more than the Mechanical Turk was. They are just getting to be better imitators.


I guess the reality is, perhaps given the current global situation, that it's better AGI doesn't exist. We are well on our way to a dystopian catastrophe, adding AGI into it when the US government is a puppet of the technocracy seems a very dangerous game.

I think people still suffer from hubris here that human intelligence is more than learning what response is appropriate then the summation of all previous responses. I think applying the definitions of AGI to humans would lead to the conclusions that humans are not an AGI.

Also that particular survey is suggesting brute force scaling won’t get to AGI seems reasonable I also don’t think there was ever an argument made that brute force scaling is the solution.

[Fuzz]

Originally Posted by GGG

I think people still suffer from hubris here that human intelligence is more than learning what response is appropriate then the summation of all previous responses. I think applying the definitions of AGI to humans would lead to the conclusions that humans are not an AGI.

Also that particular survey is suggesting brute force scaling won’t get to AGI seems reasonable I also don’t think there was ever an argument made that brute force scaling is the solution.

Musk obviously believes it, because he keeps claiming we are close. And because he is pouring all his efforts and money into LLM like systems, he must believe that's going to get them there. Unless he's lying and grifting again, which is also a strong possibility. But I think he tends to believe what he says, at least in general.


To your first point, I think we already see the progression. LLM's are bad at math, so the solution was to drop them to other systems when calculations are requested. I suspect we'll get loads of different models tweaked towards being really good at different fields and tasks. They'll all get mashed together, and a decision engine will choose which to use in which case. But they will still be an assembly of processes. Until that system can then receive a new novel field and learn the best way to solve it, they wont be AGI(or equivalent to human thinking). They'll just be a collection of systems that work in specific domains.


None of this means they are useless, but I think just being able to do most of what humans can, and in many cases far better, is not AGI anymore than a TI-82 is a brain.



I think that if we do get AGI, it'll have to be emergent that learns in it's own way on it's own. And no one has figured out how to bootstrap an emergent AI, or we'd have it. So it could be tomorrow, a decade from now, or never. But if it does come, it's going to be incredibly rapid and world changing, in ways an LLM never could.