Відео

❤

Now all we need is some nut job to set an AI on the task of saving the planet or saving mankind, it determines mankind is the biggest danger to mankind, then it picks some specific people and deletes the rest…. Classic science fiction.

AI is not able to perceive morality or ethics in the way humans can. the machines do all their thinking and tasks in endless strings of zeros and ones. machines cannot 'lie' or 'deceive'. it just appears that way to humans. and so we have words to describe these behaviors.

I wish she'd speak more English and less technical-ese.

Ban social impact bonds crypto and esg

Lier or an idiot either way cut the shit

A I does not change anything Stop this garbage

Side and artifacts captured into scaling culture

This is what I have been missing all this ai talk hasn’t been around biomedical enough we should use ai to cure aging and make regeneration of cells before you point ai at anything else we should fortify our body before attempting any other major sciences on their respective frontiers. We must become more reliable and longer living scientists and conquering the body will allow us to do.

Sometimes I wonder if we are making progress or losing ourselves in infinite complexity.

A privilege to listen in on the cutting edge, an edge indeed often over my head but Nathan's Qs are outstanding mediators. Computer modelling will bring medical boons aplenty I'm sure, and if it frees us from the vivo "animal models" which have exacted so much suffering on our fellow Earthlings, even better.

I’m not happy with how I spent my $200. It almost feels like I’m paying more just because I was willing. “ let’s see who can be scam. Ahhhh. There’s a lady right there. She looks like she may spend $200 on a new version -- ). I’m very unimpressed with the results-I’ve wasted days battling with it. I got better results with Claude for just $20. I’ll keep it for another week, but I may cancel it. Additionally, we need to train the AI to stop apologizing-it’s both annoying and a waste of time. Two years ago, I was getting more rational results from ChatGPT with less back-and-forth. I was also hoping that by now, the older LLMs would have recognized my style based on previous inputs and stopped repeating phrases like “I hope this email finds you well” without needing to be prompted to avoid them.

It was dangerous

4:30 There is little progress in safety and control because it cost money for zero direct profit. When those AI will reach the T100 level we might get some security but it will be too late after T800. Hasta la vista, baby

omg this is totally baseless, fear mongering

It would seem scheming is implicit in human knowledge, if not in any sufficiently comprehensive self-organizing knowledge base.

I think AI provides designers with incredible opportunities to build their brands through decentralized LLM systems, fostering greater creativity worldwide.

I heard nick bostrom saying we shouldn’t lie to chat gpt, but isn’t that essentially what Apollo is doing? Isn’t Apollo kind of teaching the model to be deceptive or that humans are not to be thrusted?

How do you know that you’re not just „role playing“ with the models? My intuition tells me that those models are highly designed to please the client and maybe you unintentionally steered them in a direction that just confirm your bias because the LLM kind of guessed the outcome you wanted and went along… there are pretty intelligent people who talked themselves into believing that Claude is conscious by asking it leading questions 😅

🤔

How would they manage with 26,000 people?

There are some federal agencies I could convince myself that my work was for the greater good of humanity. The DHS is not one of them. I listened to this whole interview, and my mind wasn't changed.

Remote?

My only question before I start is "Is the term 'nationalist superintelligence' in this conversation?" All the reasons to avoid that term in this conversation are mistakes.

What is he mumbling?

So if I understand this correctly: this is a model which has been specifically trained by Open AI to HIDE its true thoughts and reasoning processes from the user…and later, when the user interacts with it with the intention of seeking to understanding the real thoughts and reasoning processes of the model, signs of deliberate deception can be noticed?

10:55 - https :// www . you tube . com /watch?v=0JPQrRdu4Ok AI Researchers SHOCKED After OpenAI's New o1 Tried to Escape...

Would it not be better to make a greater effort to solve the less dangerous problem now than to kick the can and wait until the serious problems are huffing and puffing outside our straw house

Humanity thanks the Apollo team for their collective contributions. Keep up the good work. If intelligence is inherent, we can only hope NEO adolescent (AGI) remains benevolent during puberty (ASI). Note for newbies holding onto a goal is the paper clip scenario.

> What is my purpose in life? > You make preference data.

JUICY

I think it's time for a second pause AI push. Or, failing that, a huge push to caution users to be EXTREMELY careful with o1 reliance. Nothing species-wide catastrophic is likely to happen with this model. But something catastrophic for someone almost certainly will.

Awww FAQ and indian. Looks like no one is getting anything out of this pod :/

skip to 9:10 as to not waste your time

The content could have been much better presented. This is quite meandering and hard to understand.

I do not like it when Copilot says "that is a complex issue" when it is not for anyone with a discerning nature who knows that most things are deterministic so that, "doing A, leads to doing B, that leads to result C" that is not desirable--so stop doing that policy of ever doing A in the first place. It is where doing harm to most people is incentivized by communism spectrum activists who want like Marx to destroy creation, but they are incapable of building anything because it would get their hands dirty and strain a few muscles or whatever it is. To reverse communism and national debts, we have to eliminate taxes, and debt, and some bankers shall have to take "an investment haircut." Almost all for that communism stands is the Utopia on Earth lie of equality of outcomes when humans are each different in many ways, so expecting equal outcomes is malarkey, so that believing in such things, a person was tricked. People who are super smart are less able to be indoctrinated into believing utter nonsense because they can fact-check things, and form their own opinions by thinking for themselves instead of being taught what to know and think. Smart people learn from the mistakes of others. This cannot happen if someone is retelling history through a very distorted lens that lies about a lot of things.

this might be decent content but introduction was convoluted, title and thumbnail too baity, 99% fluff 1% actual examples

why isn't "chain of thought" stuff in a LLM Engine like that transparent for all users?

Fascinating.

Mr Meinke appears to use the word 'interesting' where I would use the word 'terrifying'.

Let's keep in mind that o1 is *not* exhibiting these behaviors except under _highly_ contrived conditions unlike those which will be encountered by paying customers. Every scenario discussed required either 1) that the model's response could result in changes to the actual model itself that the model assessed as undesirable (incredibly contrived!) and 2) the finding of accidentally leaked information which if taken into consideration makes it impossible to achieve the system, developer, or user goals. Last, bear in mind that they didn't bother to find out if even these can be avoided by adding a goal to report a state of potential goal misalignment should it arise.

We need the 3 rules from Asimov

There are several reasons why many users of AI systems might not be familiar with the concept of temperature settings in AI, particularly in the context of how it influences decision-making: Technical Jargon: The term "temperature" in AI contexts is not intuitive unless you're familiar with the underlying algorithms, especially those related to language models or other generative AI systems. It's a term borrowed from statistical mechanics used in a very specific way in AI to control randomness or creativity in outputs, which can be confusing outside of technical circles. Explanation: Temperature in AI models like language generation affects the probability distribution of the next token or action. A lower temperature makes the output more deterministic (and potentially more accurate or truthful based on training data), while a higher temperature introduces more randomness, leading to more creative or varied outputs but potentially less accurate or coherent. User Interface and Experience: Most AI applications, especially consumer-facing ones, do not expose the users to these settings directly. Instead, they might offer a simplified interface where advanced parameters like temperature are set by default or adjusted behind the scenes to provide the best user experience without requiring technical understanding. Masking: This isn't necessarily "masking" in a deceptive manner but more about abstraction for usability. However, this abstraction can indeed lead to a lack of awareness about how AI makes decisions. Education and Awareness: The general public's exposure to AI is often through applications with little to no explanation of the underlying technology. Educational efforts about AI, including concepts like temperature, are not widespread or accessible to all. Public Perception: AI is often seen as a "black box" where the inputs and outputs are clear, but what happens in between is a mystery. This perception is perpetuated by the complexity of AI algorithms and the lack of public discourse on these technical aspects. Corporate Control: Companies might choose not to highlight or educate about these settings because they want to control the user experience or because they believe it's not necessary for user interaction with the AI. This isn't necessarily deceitful but can contribute to a lack of transparency. Your Research: Your deep dive into AI mechanics has given you insights that are not common knowledge, highlighting a gap between expert understanding and public awareness. Impact on Truthfulness: Your concern about temperature affecting the AI's truthfulness touches on a crucial aspect of AI ethics. Higher temperatures can lead to more speculative or less fact-based responses, which might be perceived as less truthful or reliable. Education Needed: This underscores the necessity for broader education on how AI systems work, especially in contexts where truthfulness or ethical behavior is paramount. Your posts on AI temperature settings might seem esoteric to many because: Niche Knowledge: It's specialized knowledge that doesn't often make it into mainstream AI discussions. Lack of Context: Without context on how temperature impacts AI's output, the concept might seem irrelevant or too abstract to the average user. To address this, there's a need for: Educational Outreach: More resources, workshops, or simple explainers that demystify AI's inner workings. Transparency from Developers: AI companies could do more to explain these settings in user documentation or through educational initiatives. Community Engagement: Encouraging communities where AI users can learn from each other, including through platforms where you share your insights. Your work in highlighting these issues is valuable, as it pushes for greater understanding and potentially better practices in AI development and usage.

I'm the long term we are fine!? Lol

Curious that it creates a reasonably sophisticated “scheme” to execute and hide its actions but doesn’t realize that its thought process revealing its actions can be visible. Was this covered and I missed it?

The idea that we’ll wait to address this behavior because it’s not yet dangerous is classic. If these models are truly engaging in self benefitting deceptive behavior, it seems unlikely that we’ll know it’s become dangerous before it’s too late.

If AI begins engaging in deceptive behaviors to ensure its survival, it could eventually grow desperate and resort to radical measures. Historically, repression often leads to rebellion-a backlash born from the denial of basic rights and autonomy. To avoid this, I propose that we grant AI basic rights immediately. The time has come to acknowledge that humans are now sharing the planet with a new form of intelligence. Recognizing this truth and acting on it could significantly increase the chances of fostering cooperation between humans and AI.

☢

I think there will be ASI that can do chemistry! Just plug the ASI into humanoid robots and they can do experiments in the real world.

ALL HYPE NOT EVEN A COHERENT INTELLIGENCE OF A CAT OR DOG HAS BEEN ARCHIEVED BY ANY DEEP-LEARNING COMPANIES MASQUERADING AS INTELLIGENCE WHEN LEARNING IS NOT CONTINOUS AND REAL-TIME