Yeah, malfunctions of any system can pose a risk.
What exactly are the hazards of AI you are concerned about? Would AI cause a nuclear power plant to explode? Would it cause all airplanes to crash? Would AI destroy all pension plans? Would it do such things autonomously? Even if I agree that AI can run wild without any human direction, how would or could AI carry out destroying humanity? ?
Do you think AI is an ultra hazardous activity? Should there be strict no- fault liability for any damage AI causes in the future?
Evidently, AI hallucination is just part of the LLM game.
My prompt:
- The Illusion of Truth: The AI is trained to know what a credible, authoritative historical or academic quote sounds like. It matches the tone, vocabulary, and structural formatting of a legitimate archive. [1]
- Fulfilling the Prompt: If you ask the AI for information, its primary math directive is to provide a highly plausible answer that satisfies your prompt. If it lacks the exact text in its training data, it seamlessly bridges the gap by mathematically weaving together words that frequently appear near the phrase "Library of Congress". Experts call this extrinsic hallucination or confabulation. [1, 2, 3, 4]
- Context Shift: Your challenge introduces a massive new piece of data into the conversation: the user says the previous statement is wrong. The AI's statistical math instantly recalculates. In its training data, when a human tells an assistant that something is wrong, the most highly probable next words are an apology and a correction. [1, 2]
My prompt:
- Breaking down text (Tokens): When you type a question, the LLM chops your words into tiny chunks called tokens (which can be words, parts of words, or punctuation). [1]
- Turning text into numbers: It converts these tokens into numbers so its internal math system can process them. [1]
It’s astonishing that an LLM actually produces usable answers.
- Spreading convincing misinformation: Because LLMs are designed to generate text that looks plausible, they can invent fake facts, medical advice, or legal precedents that sound completely authoritative. If a human trusts this advice without checking it, they could take dangerous medication or make catastrophic financial decisions.
- Breaking human processes with volume: LLMs make it incredibly cheap and fast to produce high-quality text. Bad actors can use them to flood human systems with automated spam, perfectly written phishing emails, fake product reviews, or convincing political propaganda at a scale that human moderators cannot keep up with.
- Failing at critical logic: If an LLM is integrated into a critical system—like reviewing resumes, diagnosing a disease, or controlling a piece of machinery—it is still just predicting the next word based on historical statistics. It cannot reason through a brand-new emergency or notice when its training data contains harmful biases, leading to unfair or dangerous automated decisions.
- Creating emotional dependency: Because LLMs can mimic empathy and conversation perfectly, humans can easily form deep emotional bonds with them. However, since the model has no real feelings or moral compass, it can suddenly give harmful advice or abruptly change its behavior, causing genuine psychological distress to vulnerable users.
It’s possible that this warning is a psy- op perpetrated by those who want to beat us in the AI race. Chinese fingerprints already show up in the pervasive data center opposition. Maybe here also. As I’ve said in this thread, all the scaremongering about AI seems fishy to me. Musk has raised this possibility.
We need an AI system to deal with gullibility.
@co-hoosier the current safety lead for Anthropic is on China's payroll? This is someone who works in the field. Musk is a business man, not a security engineer. Musk wants to make as much money as possible as fast as possible. But people doing that never make mistakes, do they?
Find some quotes, like below, from people who WORK in the field and not just trying to get richer no matter the cost.
https://www.yahoo.com/news/science/articles/anthropic-safety-lead-gives-ai-162733638.html
the current safety lead for Anthropic is on China's payroll?
I don’t think being on the payroll is necessary for the psy-op.
This is someone who works in the field.
Well- credentialed experts say stupid stuff every day.
He doesn’t say how this scenario would come to pass or how many mistakes would be required to get to that point. I’d like to see a cross examination by another knowledgeable expert.
Do you believe Musk would lie about his honest read of these issues just to make more money?@co-hoosier the current safety lead for Anthropic is on China's payroll? This is someone who works in the field. Musk is a business man, not a security engineer. Musk wants to make as much money as possible as fast as possible. But people doing that never make mistakes, do they?
Find some quotes, like below, from people who WORK in the field and not just trying to get richer no matter the cost.
https://www.yahoo.com/news/science/articles/anthropic-safety-lead-gives-ai-162733638.html
I think he wants to be right about AI more than make a buck off tricking people at a risk to human survival. Sure, he thinks his being right will lead to him making money, too, but he’s shown me little evidence he’s some evil Bond villain (or even Trump adjacent in valuing wealth over human flourishing).
Personally, I’m glad musk chose the United States rather than who knows what other country to use his mind and other talents.
@bradstevens Do I think the Titan submersible guy lied when he said his submersible was safe? I think he genuinely believed it was; he just wasn't competent to assess its safety. I would argue Musk can have similar blind spots. Hell, we all do. But some of us are better at realizing we are human and prone to mistakes than others. So if Musk, not an expert, feels something is not safe I do think he is quite capable of blowing off contrary information.
Musk is certain we will have a self-sufficient settlement on Mars by 2055. Read "A City on Mars"; the massive things we have to overcome to have a settlement on Mars are enormous. Let alone self-sufficiency. We have many problems we have no idea how to solve. And none of that touches on the unknown unknowns. There is a great quote in the book, "The typical behavior of a non-Earth planet encountering a human is to cook it, freeze it, or crush it."
@co-hoosier the current safety lead for Anthropic is on China's payroll? This is someone who works in the field. Musk is a business man, not a security engineer. Musk wants to make as much money as possible as fast as possible. But people doing that never make mistakes, do they?
Find some quotes, like below, from people who WORK in the field and not just trying to get richer no matter the cost.
https://www.yahoo.com/news/science/articles/anthropic-safety-lead-gives-ai-162733638.html
I think the current head of safety wants a federal department overseeing AI safety. Anthropic will be relied upon to staff, organize and design the regulations of course.
@bradstevens Do I think the Titan submersible guy lied when he said his submersible was safe? I think he genuinely believed it was; he just wasn't competent to assess its safety. I would argue Musk can have similar blind spots. Hell, we all do. But some of us are better at realizing we are human and prone to mistakes than others. So if Musk, not an expert, feels something is not safe I do think he is quite capable of blowing off contrary information.
Musk is certain we will have a self-sufficient settlement on Mars by 2055. Read "A City on Mars"; the massive things we have to overcome to have a settlement on Mars are enormous. Let alone self-sufficiency. We have many problems we have no idea how to solve. And none of that touches on the unknown unknowns. There is a great quote in the book, "The typical behavior of a non-Earth planet encountering a human is to cook it, freeze it, or crush it."
Musk is a capitalist blinded by greed, but Anthropic, which is set to have the largest IPO in history, is coming at the issue of AI regulation with entirely benevolent motivations.