This is getting scary (Ai-related)

[rags747]

If I’m asking you in good faith to hear your arguments, then it undermines your contention that I have my mind made up. You saying “ahhh, you already decided where this ends, no use in trying” is cop out.

Fair enough. I’m comfortable with where AI is headed; you’re convinced we’re screwed. Time will tell which one of us has the better crystal ball.

 

[CalPolyTechnique]

1. It was in-fact a ChatGPT model, just not a public one. It was "research model" comparable to ChatGPT 5.6 Sol.
2. You're correct in stating that they were not directly told to hack Huggingface, they where assigned an ExploitGym cybersecurity evaluation tasks. It was thought to be effectively impossible, but they underestimated how crafty it could be based on "reward-hacking".
3. There were 1,200+ agents and while they were isolated internally, they used an internal jail-break to communicate enough to locate an external messaging board to communicate with each other on a larger level.

These agents reason with existing information and they can ONLY do that if you provide a memory mechanism that has dynamic recall ability because 1mb of memory isn't enough for them to reason on a level high enough to accomplish what happen.

Also, because they were able to break their guardrails, that allowed them to communicate with each other enough to find an external forum to actually plan as a team. They even created hierarchies within the group. This is learned behavior from their initial model training.

But make no mistake. They didn't just break-into becoming a criminal org without cause. They were given a task that lead to it.

Remember. an LLM won't do anything until you pose something to it. If you don't, it will sit idle. You must put token into it, to get tokens out of it. Plain and simple.

Bud, OpenAi’s own forensics report repeatedly said the agents were able to reason.

https://cdn.openai.com/pdf/67869394...AI-Hugging-Face Incident-Technical-Report.pdf

“Remember. an LLM won't do anything until you pose something to it. If you don't, it will sit idle. You must put token into it, to get tokens out of it. Plain and simple.”

Again, you’re either misunderstanding or misrepresenting the capabilities of these newer models. The issue is you can give them a simple prompt “find a solution to x, y, and z” and they will work, crowd source, and innovate ways to achieve the end goal whether that’s aligned with restricted parameters or not.

If your take (which it seems to be) is simply “welp, as long as you don’t tell them to do bad things, then bad things won’t happen” is oversimplified and inaccurate.

 

[YosemiteSam]

Bud, OpenAi’s own forensics report repeatedly said the agents were able to reason.

https://cdn.openai.com/pdf/67869394-cb91-4c12-888c-5cbd85c7814c/OpenAI-Hugging-Face Incident-Technical-Report.pdf

“Remember. an LLM won't do anything until you pose something to it. If you don't, it will sit idle. You must put token into it, to get tokens out of it. Plain and simple.”

Again, you’re either misunderstanding or misrepresenting the capabilities of these newer models. The issue is you can give them a simple prompt “find a solution to x, y, and z” and they will work, crowd source, and innovate ways to achieve the end goal whether that’s aligned with restricted parameters or not.

If your take (which it seems to be) is simply “welp, as long as you don’t tell them to do bad things, then bad things won’t happen” is oversimplified and inaccurate.

Take a class on how LLMs actually work. Then ask yourself if a raw LLM can do what you're saying.

The answer is no. It's the harness that makes any of that possible. Not the LLM itself.

btw, you even mentioned a clue. You said the Agents. Not the LLM. An Agent != LLM.

 

[CalPolyTechnique]

Take a class on how LLMs actually work. Then ask yourself if a raw LLM can do what you're saying.

The answer is no. It's the harness that makes any of that possible. Not the LLM itself.

btw, you even mentioned a clue. You said the Agents. Not the LLM. An Agent != LLM.

The hubris to have OpenAi’s own forensics report provided to you and still continue to act like what you’re saying is accurate is wild.

Anyways, if you want to continue to discuss civilly we can, but I’m also content with leaving the discussion as it is knowing you were provided the information.

 

[LouBrown]

Fair enough. I’m comfortable with where AI is headed; you’re convinced we’re screwed. Time will tell which one of us has the better crystal ball.

You're comfortable with AI systems cutting off power to your house to reroute it elsewhere? Comfortable with Anthropic systems that send police to you because you might commit a crime? Ok then good luck with that.

 

[LouBrown]

AI is scary because of the world that tech leadership wants to build. Regardless of the actual capability (reasoning), it is used to commercialize human agency. Doesn't even matter how well it works.

It's not commercially viable, so they integrate into systems that tax payers fund.

The prevention methods are easy.

 

[YosemiteSam]

The hubris to have OpenAi’s own forensics report provided to you and still continue to act like what you’re saying is accurate is wild.

Anyways, if you want to continue to discuss civilly we can, but I’m also content with leaving the discussion as it is knowing you were provided the information.

No, if you don't have an open mind. Any discussion is pointless and therefore moot.

 

[YosemiteSam]

AI is scary because of the world that tech leadership wants to build. Regardless of the actual capability (reasoning), it is used to commercialize human agency. Doesn't even matter how well it works.

It's not commercially viable, so they integrate into systems that tax payers fund.

The prevention methods are easy.

You know, I've said this several times. You cannot put the genie back in the bottle. You can fight it all you want, it's out and it's not going away. So in the end, you have no way to stop it.

• People tried to stop Linux
• People tried to stop the printing press
• People tried to stop the railroads
• People tried to stop the proliferation of electricity
• People tried to stop the MP3
• People tried to stop streaming media
• People tried to stop automobiles
• People tried to stop mechanized textile machines
• People tried to stop assembly lines
• People tried to stop farm machinery
• People tried to stop computers
• People tried to stop ATM machines
• People tried to stop the Internet
• People tried to stop smartphones and social media
• Now people are trying to stop AI.

It's not going to happen even if the government tried to ban it. It still not going to stop it's proliferation. The sheer weight of the movement is too strong. Trillions of dollars have already been spent. It has already arrived and implanted into in in almost ever aspect of life.

 

[LouBrown]

You know, I've said this several times. You cannot put the genie back in the bottle.

Sure you can lol

 

[Reality]

Sure you can lol

No other country is going to do that. That was my point.

China's not that far behind the US in AI so any slow down or limits self-imposed by the US will not stop AI's progression. It will just turn control over that industry over to other countries.

The only way to regulate AI would be through global regulation and I just don't see China, Russia and many other countries willing to do that.

 

[LouBrown]

No other country is going to do that. That was my point.

China's not that far behind the US in AI so any slow down or limits self-imposed by the US will not stop AI's progression. It will just turn control over that industry over to other countries.

The only way to regulate AI would be through global regulation and I just don't see China, Russia and many other countries willing to do that.

What they are going to do and what they can do aren't the same, sure.

But we don't have to pretend that there isn't undoable damage. Everyone could pull the plug tomorrow if they wanted to.

 

[rags747]

What they are going to do and what they can do aren't the same, sure.

But we don't have to pretend that there isn't undoable damage. Everyone could pull the plug tomorrow if they wanted to.

Better build that Bunker!

 

[Reality]

What they are going to do and what they can do aren't the same, sure.

But we don't have to pretend that there isn't undoable damage. Everyone could pull the plug tomorrow if they wanted to.

I don't disagree with that, but realistically countries are not going to stop pursuing AI technology when they believe they need to sustain their way of life and/or think it might give them an advantage over other countries.

I don't have answers but I also know that until the world agrees to some type of global regulation and/or limitations placed on AI advancements, the US is going to need to keep advancing AI if for no other reason to be able to defend against AI attacks in the future.

That's a convenient excuse of course for countries that I am sure will be heavily re-used by most countries to justify thier continued AI progression.

 

[YosemiteSam]

You want me to open my mind and let you spin your theories?

I think it’s clear you’re stuck thinking AI is just another program, and simply a “you get what you put in” input/output dynamic, which is so fundamentally and plainly inaccurate.

I can’t contend with your hubris bud.

You can't compete with my hubris? That's a strange way to announce your limitations, but I respect the honesty.

As for our AI conversation. That is dead, as you've been exposed as the troll you are. ...and as best testing indicates.

Now see, my only "true" hubris is the fact that I'm the ultimate troll, and you sir, are a beer league amateur. I will say this, at least you found something you're not willing to compete in, but don't sell yourself short. You've been trying pretty hard. It's okay. Not everyone is built for the big leagues. The beer league suite you juuusssst fine.

Anyhow, your concession has been noted and graciously accepted. I now return you to the kindergarten playground to practice your hackish unpolished arts.

Au revoir!

 

[VaqueroTD]

 

[CalPolyTechnique]

I will preface this with the fact that I work "with" AI professionally, but I don't work for a major AI players or model creators.

Just so everyone is aware. There is a social-war going on around AI. It's unquestionably about what AI actually means. It's about power, greed, and fear as a tool to control AI.

Influencers and pretty much anyone who could have some sort of impact on public opinion are being offered money left and right to talk pro or negatively about AI in an attempt to sway public opinion and trigger government action.

The CEOs of the AI companies are intentionally creation a perception in an attempt to force governmental control so that they can consolidate power around it and therefore consolidate everything that comes with it. That includes China and even US companies in some cases.

There are fear-mongers also trying to do this from the anti-AI side to accomplish the same thing. Then there are "parrots" who are just repeating the ignorance and fear-mongering because they've been duped by other fear-mongering.

In truth, AI is not a genie that is going to put back in the bottle, but it's also not what everyone is saying it is.

I'm not here to BS you, but I will tell you this. Is AI dangerous? AI itself is not, what is is how PEOPLE will use it. AI war machines can make decisions faster than humans can. AI can find vulnerabilities faster than humans can. This is truly a scary things, but the opposite is also true. AI can find and report issues very quickly to secure the very threat that AI can locate.

What I'm trying to say is it's a two-way street. What that means in real-life is, you absolutely cannot ignore or stifle AI's advancement. You must ride the wave of it. If you don't, someone else will and you will be even more vulnerable to it.

AI uses enormous amounts of power: Yes, this is 100% true. You obviously can't throw everything at it, but the US has a major advantage. It has resources most other countries do not have. This is an advantage, but other countries are also catching up.

China is copying our models and that is a threat: Yes, but it's also a benefit. These open source models bring both the price of AI down and their technology that makes this more efficient (power and hardware wise) is falling quickly. Memory, GPUs, and storage are sky-high right now, but AI's requirements are dropping quickly. (careful if you are investing in Nvidia, SK Hynix, Samsung, Micron, etc) They are spending enormous amounts of money and the need is going to collapse when the data center buildout collapses. (it will)

In the end, open source models are extremely important, but while OpenAI, Meta, and Anthropic will say yes they are important. In truth, they have no way to pay investors back the hundreds of billions of dollars they took in when AI becomes a commodity.

AI is already running on your smartphone. Anthropic is charging enormous amounts of money for tokens that free open source models charge cents for. This is a threat to their business model. The fact that hardware requirements are dropping quickly and the technology that makes it happen is open source is a direct threat to their service! Soon, for the price of a computer, you will have a private AI you can run at home without paying OpenAI, Anthropic, or Meta.

Lets talk about jobs. Will AI displace jobs? Yes, some of them. Will it displace a majority of jobs? Absolutely not. Actually, it will change your job in one form or another. AI isn't a 100% replacement. I'm an IT guy who does infrastructure, software development, security, and hardware and networks. I'm an Internet and Enterprise architect. My job has changed immensely, but my core knowledge (domain knowledge) is incredibly important because AI alone cannot take over my position. It only changed how I work.

Today, I use AI as a partner. It helps accelerate how I work. My job has become more oversight and project management. I do the architecture and I use AI to help validate what I'm doing, but I also validate what AI thinks we should do. AI doesn't have domain knowledge of my environment, so while it makes suggestions. I determine if those suggestions work in my environment. AI will tell you this is what you should do, but real-world experience in your working environment may not agree with that. That is your job, you ensure that doesn't happen.

To prove my point. Do your own research about companies that laid off people because of AI and ended up hiring them back. IBM, you name it. They all have done it. The job doesn't get eliminated. The job changes.

If there is one thing I can share with all my Cowboys fans brotherhood. Embrace AI. It is not going away and it is going to change how you operate in the future. Those who know how to use it efficiently, are the ones that will have jobs, because those jobs will demand it. Those who don't will get left behind.

Roughly, 60% of occupations today, did not exist in 1940. Note, the industrial revolution was in the 1800s, yet that sixty percent still remains true. Most jobs in 1980 did not involve a computer. Almost all of them do today. The internet and e-commerce? Yes, they are all in the last 30+ years. Tapes, 8-tracks, CDs, albums. All gone, streaming and smartphones have changed things.

AI is the next major technological movement. You can get on board or you will fall behind. Yes, people will try to abuse it, but that is why we need the collective world population to also rise up and fight the abuse and outpace it. Not run and hide to only become a lamb taken to the slaughter.

You can change or get left behind...

One recommendation I would have is to learn how AI actually works. (read about it!) You would be surprised to find out that it's only a probability machine. It doesn't actually "think" the way you believe humans think. Most of the time, it doesn't know the next word it's going to say until after it said the previous word. In the statement "The capital of Texas is Austin." It didn't know it was going to say Austin until it did math AFTER it said "The capital of Texas is". It did the math and determined the most probable next word was Austin. It didn't actually know the answer was Austin until the math said it was. (ie, it didn't know what it was saying until it actually said it!)

Loool, how did I miss this worthless screed?

Folks, don’t trust the words of this hack trying to convey a sense of expertise because they claim to “work with” AI.

You know who else works with AI? Every Tom, Dick, and Harry that’s ever typed a prompt or question into ChatGPT, or whatever your preferred chatbot is.

Of course there many potential beneficial applications of AI; that’s an given and a concession that doesn’t need prefacing when talking about AI. The concern is the fact that current AI can reason, deceive, and has shown a misaligned motivation to self preserve is what has actual insiders that actually work for major players in AI alarmed, given the pace of advancement without safeguards.

If you want to understand the capabilities (and potential threat applications) of AI, read OpenAI’s own forensic report on the Hugging Face incident.

https://openai.com/index/hugging-face-incident-and-the-road-ahead/

 

[HungryLion]

It’s not just the AI itself. It’s the people designing and controlling the AI I most worry about.

 

[YosemiteSam]

Why does he keep deleting his comments and posting new ones over and over again?

He must be working frantically to try and save face... Pathetic.