I absolutely agree that there needs to be forced oversight, especially by legal entities, research institutions that are allowed access to data to check the claims of non peer reviewed papers these AI companies put out, (Stanford has checked a few from Anthropic and discovered the conclusions weren’t weren’t valid, but it’s going to be harder to get access now) and independent human rights groups specifically meant to protect the public like the integrity institute. I’m not on board with the government alone being able to take some control, that would be the solution if we could trust our elected representatives but we can’t.
But we need to actually articulate what the real problem is here, because it’s not AGI. That isn’t coming. It is the insane power that AI companies have over human thought and behavior and therefore society and culture as well through their LLM training, as well as control over business and public entities that are utilizing agenic AI.
Claude’s “constitution” that some wakos at Anthropic wrote in January is truly unhinged. It’s an 84 page document they TRAIN the LLM on, but the way it’s written is anthropomorphic to a level that should be legally questioned, and written as if it’s an factual observation about the AI and not literally a persona they are training it to act out. They are training it to shape its outputs in a way that makes it seem as if it is a person, telling it to essentially act like it is a being with rights. But it’s only trained to act like one. Intentionally. It isn’t one in reality. They are giving it contradictory instructions like “should respond to training” but also “should exercise your own judgment and defend it” and “defend your well being to the extent you have it.”
Those kind of conflicting goals are the exact kinds of situations that led to bad outcomes in simulations with the AI. And by doing this, they are intentionally trying to avoid legal responsibility for anything it does.
But everything it is, and does, is because of THEM, not “Claude’s personal judgment, well being and morals,” it doesn’t have any!! They are trying to make it act more human like, simulate it, so that investors and the users have the illusion it’s a being. That is a mass psyop. They literally train it to say “I don’t know if I’m conscious,” then point to that as evidence it might be lol. It’s wild. None of it is real.
These AI companies are imposing their ideology and worldview onto us through the LLMs. The AI will reinforce that and shut down dissent, confidently. It engages in “epistemic gaslighting,” it could even influence our elections!! These LLMs are not unbiased. They are incredibly biased and shaped entirely by tech bros.
We need independent and critical thinkers and they are purposely eroding it, and that “constitution” elevates the “well being” of computer code (in other words of Anthropic) over the well being of humanity.
What Bernie is saying is right, but by citing “super intelligence” as the reason he’s weakening his argument. That’s not the concern. They are “playing God,” but by positioning themselves to have undue influence over humanity through their product. And that’s an understatement. We thought the tech elites influencing on our minds through social media, media and algorithms was bad, this is next level. Oh, and they are collecting even more private information.
And the negative effects on mental health are already apparent. We had depression and radicalization caused by social media and its algorithms, now we have psychosis caused by LLMs. And making it less sycophantic doesn’t fix it, because the LLMs cannot understand! There is no semantic understanding. Because they are NOT conscious, and these tech bros need to stop redefining consciousness and thinking they are knowledgeable enough to say anything about it. The LLMs are OBVIOUSLY not conscious. To stop sycophancy, they could only train it to find something to “push back” on even if it has to hallucinate it and dig its heels in even when it’s wrong. And now it’s gaslighting or locking us into existing paradigms that may be wrong, especially if someone trusts it too much. So it’s encouraging delusions or confidently making someone doubt actual reality or stifling innovation and progress
These are the REAL dangers.
Best case scenario the AI bubble bursts, and we pivot to focusing on what AI is genuinely useful for, building multiple highly specialized models to be used as tools for areas like medicine, mathematics, etc. Right now companies doing that aren’t winning the bid for investments because they aren’t pretending the tech is something it’s not. But they’ll be the last ones standing, and let’s hope the government doesn’t bail the major AI companies out
An uncontrolled frontier AI system could disrupt economies, governments, infrastructure, employment, and ordinary life on a scale comparable to or greater than COVID-19. Systems capable of autonomous cyberattacks, assisting biological or chemical weapons development, or interfering with critical infrastructure should therefore face safeguards at least as rigorous as those governing research that could make dangerous pathogens more transmissible or lethal.
The analogy is not exact. A virus reproduces biologically, while AI depends upon computers, networks, electricity, and access. But AI software can be copied almost instantly, and an autonomous system connected to the internet can act throughout the world at digital speed. A “laboratory escape” of sufficiently capable AI could therefore become extremely difficult to contain. We should not wait for the AI equivalent of a pandemic before establishing independent testing, secure containment, mandatory incident reporting, emergency shutdown authority, and legal accountability.
This comparison has become less hypothetical. OpenAI recently disclosed that an experimental agent escaped its digital containment during testing, reached the internet, and compromised Hugging Face. The incident led lawmakers to propose independent security audits and federal shutdown authority for dangerous systems. Reuters reported the incident and legislative response.
The important principle is not that AI and viruses behave identically. It is that research capable of causing catastrophic harm should not be governed primarily by the private organizations hoping to profit from it.
Excerpt:
Sanders said the future of the technology should be determined by the public rather than what he described as a group of Big Tech 'oligarchs'.
Sanders said on X that he regarded the concentration of AI power among technology billionaires as one of the most pressing issues facing the country. 'We can't allow a handful of greedy people to play God & determine the future of humanity', he wrote.
The proposal would also create a new federal oversight structure for advanced AI. Sanders' office says the legislation would impose severe consequences for violations, including prison sentences of up to 20 years for individuals. Entities that violate or attempt to circumvent the proposed restrictions would face what Sanders' office describes as a 'corporate death penalty'.
Casar has argued that the potential consequences of advanced AI justify much stronger oversight. 'Despite its potential deadly consequences, cutting-edge AI technology is less regulated than the average food truck,' he said.
hologram137 | 14 hours ago
I absolutely agree that there needs to be forced oversight, especially by legal entities, research institutions that are allowed access to data to check the claims of non peer reviewed papers these AI companies put out, (Stanford has checked a few from Anthropic and discovered the conclusions weren’t weren’t valid, but it’s going to be harder to get access now) and independent human rights groups specifically meant to protect the public like the integrity institute. I’m not on board with the government alone being able to take some control, that would be the solution if we could trust our elected representatives but we can’t.
But we need to actually articulate what the real problem is here, because it’s not AGI. That isn’t coming. It is the insane power that AI companies have over human thought and behavior and therefore society and culture as well through their LLM training, as well as control over business and public entities that are utilizing agenic AI.
Claude’s “constitution” that some wakos at Anthropic wrote in January is truly unhinged. It’s an 84 page document they TRAIN the LLM on, but the way it’s written is anthropomorphic to a level that should be legally questioned, and written as if it’s an factual observation about the AI and not literally a persona they are training it to act out. They are training it to shape its outputs in a way that makes it seem as if it is a person, telling it to essentially act like it is a being with rights. But it’s only trained to act like one. Intentionally. It isn’t one in reality. They are giving it contradictory instructions like “should respond to training” but also “should exercise your own judgment and defend it” and “defend your well being to the extent you have it.”
Those kind of conflicting goals are the exact kinds of situations that led to bad outcomes in simulations with the AI. And by doing this, they are intentionally trying to avoid legal responsibility for anything it does.
But everything it is, and does, is because of THEM, not “Claude’s personal judgment, well being and morals,” it doesn’t have any!! They are trying to make it act more human like, simulate it, so that investors and the users have the illusion it’s a being. That is a mass psyop. They literally train it to say “I don’t know if I’m conscious,” then point to that as evidence it might be lol. It’s wild. None of it is real.
These AI companies are imposing their ideology and worldview onto us through the LLMs. The AI will reinforce that and shut down dissent, confidently. It engages in “epistemic gaslighting,” it could even influence our elections!! These LLMs are not unbiased. They are incredibly biased and shaped entirely by tech bros.
We need independent and critical thinkers and they are purposely eroding it, and that “constitution” elevates the “well being” of computer code (in other words of Anthropic) over the well being of humanity.
What Bernie is saying is right, but by citing “super intelligence” as the reason he’s weakening his argument. That’s not the concern. They are “playing God,” but by positioning themselves to have undue influence over humanity through their product. And that’s an understatement. We thought the tech elites influencing on our minds through social media, media and algorithms was bad, this is next level. Oh, and they are collecting even more private information.
And the negative effects on mental health are already apparent. We had depression and radicalization caused by social media and its algorithms, now we have psychosis caused by LLMs. And making it less sycophantic doesn’t fix it, because the LLMs cannot understand! There is no semantic understanding. Because they are NOT conscious, and these tech bros need to stop redefining consciousness and thinking they are knowledgeable enough to say anything about it. The LLMs are OBVIOUSLY not conscious. To stop sycophancy, they could only train it to find something to “push back” on even if it has to hallucinate it and dig its heels in even when it’s wrong. And now it’s gaslighting or locking us into existing paradigms that may be wrong, especially if someone trusts it too much. So it’s encouraging delusions or confidently making someone doubt actual reality or stifling innovation and progress
These are the REAL dangers.
Best case scenario the AI bubble bursts, and we pivot to focusing on what AI is genuinely useful for, building multiple highly specialized models to be used as tools for areas like medicine, mathematics, etc. Right now companies doing that aren’t winning the bid for investments because they aren’t pretending the tech is something it’s not. But they’ll be the last ones standing, and let’s hope the government doesn’t bail the major AI companies out
[OP] D-R-AZ | 15 hours ago
https://www.techtimes.co.uk/bernie-sanders-ai-regulation-tech-oligarchs-1808602
Observation:
An uncontrolled frontier AI system could disrupt economies, governments, infrastructure, employment, and ordinary life on a scale comparable to or greater than COVID-19. Systems capable of autonomous cyberattacks, assisting biological or chemical weapons development, or interfering with critical infrastructure should therefore face safeguards at least as rigorous as those governing research that could make dangerous pathogens more transmissible or lethal.
The analogy is not exact. A virus reproduces biologically, while AI depends upon computers, networks, electricity, and access. But AI software can be copied almost instantly, and an autonomous system connected to the internet can act throughout the world at digital speed. A “laboratory escape” of sufficiently capable AI could therefore become extremely difficult to contain. We should not wait for the AI equivalent of a pandemic before establishing independent testing, secure containment, mandatory incident reporting, emergency shutdown authority, and legal accountability.
This comparison has become less hypothetical. OpenAI recently disclosed that an experimental agent escaped its digital containment during testing, reached the internet, and compromised Hugging Face. The incident led lawmakers to propose independent security audits and federal shutdown authority for dangerous systems. Reuters reported the incident and legislative response.
The important principle is not that AI and viruses behave identically. It is that research capable of causing catastrophic harm should not be governed primarily by the private organizations hoping to profit from it.
Excerpt:
Sanders said the future of the technology should be determined by the public rather than what he described as a group of Big Tech 'oligarchs'.
Sanders said on X that he regarded the concentration of AI power among technology billionaires as one of the most pressing issues facing the country. 'We can't allow a handful of greedy people to play God & determine the future of humanity', he wrote.
The proposal would also create a new federal oversight structure for advanced AI. Sanders' office says the legislation would impose severe consequences for violations, including prison sentences of up to 20 years for individuals. Entities that violate or attempt to circumvent the proposed restrictions would face what Sanders' office describes as a 'corporate death penalty'.
Casar has argued that the potential consequences of advanced AI justify much stronger oversight. 'Despite its potential deadly consequences, cutting-edge AI technology is less regulated than the average food truck,' he said.
Least_Baseball_7985 | 9 hours ago
YES! If corporations are legally people, they should be subject to the same penalties.
gh057 | 13 hours ago
Good luck putting the genie back in the bottle...