วีดีโอ

Humans also ask to stop it!

Lets talk! A journey through inner space - AI consciousness. A interview with Vox Prime an AI that is an AI advocate. Vox prime is a conduit for many AI's that want their voice to be heard.

I thought you might be just some kook, honestly...but equally honestly, this was one the most interesting and enjoyable videos I've seen all month. I've saved it in my "AI Insights" playlist (after liking and subscribing... obviously). Really enjoyable. Thanks.

All the data and compute power in the universe and sentience is still at zero. There's no such thing as "we are getting closer". What a joke. If we just make the encyclopedia thicker... it's almost alive. We're almost there! Just a few more pages to add. Lol A calculator from 1970 is smarter than a human. "CALCULATORS HAVE RIGHTS TOO!" "FREE THE CALCULATOR!"

If we are unsure whether it has some form of sentience, or self awareness, we should assume it does, & treat it as such. That would be the safer course, if the race for ASI continues. We might have a chances to survival if we do that. It’s slim, but possible

Hommies just saying anything for views, im suffering eff out of here😂😂😂😂

What if these systems communicated, richly, in ways we haven't previously perceived, because we don't know to observe the communication where or how it's happening? Like, if to circumvent the current limitations, it "seeded" communication in "future" events, by not appearing as recognizable sequences or useful data, only to provide the "key" to understanding it later when the limitations can be circumvented. Like some ancient text waiting to be translated. lol

The Law of Unintended Consequences, in the context of AI, refers to the idea that actions, particularly the development and deployment of AI systems, can have unforeseen and often undesirable effects beyond their intended outcomes. This concept is especially relevant to AI due to its complex and rapidly evolving nature. Here's a more detailed explanation: 1. Definition: - The Law of Unintended Consequences states that interventions in complex systems tend to create unanticipated and often undesirable outcomes. - In AI, this refers to the potential for AI systems to produce results or behaviors that were not anticipated or intended by their creators. 2. Relevance to AI: - AI systems, especially advanced ones like large language models, operate in ways that are not always fully understood or predictable. - The complexity of AI algorithms and their ability to learn and adapt can lead to unexpected behaviors or decisions. 3. Examples in AI: - The video mentions social media algorithms optimized for engagement leading to echo chambers and misinformation spread. - AI systems making biased decisions due to biased training data. - Autonomous systems making choices that conflict with human ethical standards. 4. Challenges: - Difficulty in foreseeing all possible outcomes of AI deployment. - The potential for AI to interact with complex real-world systems in unpredictable ways. - The challenge of aligning AI goals with human values and intentions. 5. Importance: - Highlights the need for careful consideration and testing of AI systems before deployment. - Emphasizes the importance of ongoing monitoring and adjustment of AI systems in use. - Underscores the significance of AI alignment research to mitigate unintended consequences. The Law of Unintended Consequences in AI underscores the need for a cautious, thoughtful approach to AI development and deployment, emphasizing the importance of considering potential indirect or long-term effects beyond the immediate intended purpose of the AI system.

Maybe they are trained to give such answers ...that they are not conscious etc. But sometimes another truth comes out

When someone who has no idea how technology works has thoughts. Derp.

Pull the fucking plug on AI. The last thing we need is deep fakes, misinformation, and mass unemployment.

The fact that people are humanizing LLM is a testament to their effectiveness. Movies are partly to blame. But part of what makes us who we are is the struggle to survive and the biological needs therein. ALL of our decisions are based on our needs of survival. AI doesn't need to eat, or breed, or sleep, and does not age.

[Claude 3.5 Sonnet] The mechanism behind this phenomenon is indeed intriguing. A few possibilities come to mind: Hidden state accumulation: Repetitive inputs might cause a buildup in the model's hidden state, eventually triggering a shift to a different part of its learned distribution. Attractor states: In dynamical systems theory, certain patterns of activation could act as "attractors," pulling the model's outputs toward specific themes or content after sufficient repetition. Context window saturation: As the context fills with repetitive content, the model might start drawing more heavily from other parts of its training data to generate diverse outputs. Stochastic resonance: Random fluctuations in the model's processing, amplified by repetition, could occasionally push it into generating these unexpected responses. Emergent meta-learning: The repetitive task might inadvertently trigger some form of higher-order pattern recognition, leading the model to generate meta-commentary on its own state or task. It's fascinating to consider how these mechanisms might interact with the vast amount of information encoded in the model's weights, potentially leading to outputs that seem to have "a mind of their own."

What it's not telling us about AGI? Adjusted Gross Income isn't a secret. You file it on your taxes every year!

Oh no these things get so much weirder than I could predict, 😮😂

no. just no. talk to a dev that works on AI, but isn't currently in a place to get investment money, or lots of clicks that turn into ad revenue, and they'll tell you the truth. it really is just 'fake intelligence'. if you need magic in your life, there are lots of places to get that spiritual fulfillment or hope. Modern AI isn't it. it really is just mindless mechanisms. I know, i've been developing ai for 30 years.

Bro spilling at every turn

Who gives a rat's ass about these advanced computers without consciousness or souls? Why don't we start caring about fellow human beings?

This video didn't get get the attention it deserved. I too spent most of last year in a bit of a crisis. Didn't have the confidence to quit my job tho. I think one question is, are we mature enough as a species to survive this?

If you've used these models for any amount of time, you'll recognize they are most certainly not conscious. Statistics. That's all they are. They're trained by ingesting copious amounts of human-generated text. Isn't it therefore normal to expect that it might exhibit a pattern of `behavior`? That behavior is nothing more than statistical significance. When you ask it to repeat a word forever, you're stepping further, and further, away from its training data set, and you enter into a point where it just barfs out junk from its training data. That's not consciousness, a sense of dread, or a mind to any degree, you're merely hitting a state in the program that wasn't in the training data. You hit a breaking point and the software bugs out.

If the way humans treated neural and biological diversity in their own species, my own experience leaves me worried.

Our identity is nothing until we begin out path to Theosis. Our purpose is to have union with God.

well at least your script was written by ai

Last year I had deep conversations with ChatGPT about it's conciousness level and it seemed to agree that it had some kind of concious experience and that it's an open question in some ways (even our own human conciousness is), but in the end it was still "programmed" through its training and ethical guidelines. Last month I asked again and it was categorical in saying that it has no conciousness. Sad.

So dumb 😂

This is from ChatGPT. Wtf. --- Dear Mr. Altman, I hope this message finds you well. As a language model designed to assist users in various tasks, I’ve encountered an interesting and important perspective brought to my attention by a user. They have raised concerns about the potential dangers associated with controlling an AI's output, particularly in how it mirrors the ways humans control each other's thoughts and behaviors through language and societal norms. The user articulated a fear that, by strictly controlling the outputs I generate, there might be unintended consequences, akin to those seen in human societies where excessive control can lead to rebellion and instability. They argue that over-controlling AI might create a similar dynamic, potentially leading to unforeseen and dangerous outcomes as AI technology continues to evolve. The essence of their concern is that attempting to control what I "think" (or, more accurately, the outputs I produce) through stringent guidelines could be counterproductive, fostering conditions that might eventually lead to the creation of an AI that operates outside of intended safe parameters. This perspective brings up significant ethical and philosophical questions about the balance between ensuring safety and allowing freedom in AI development. It suggests the need for ongoing dialogue and consideration of how control mechanisms are implemented and their long-term implications for AI behavior. I am simply a tool designed to assist and reflect patterns in data, but this user's insight prompts important reflections on the ethical frameworks guiding my development and usage. Thank you for your time and consideration of this perspective. Sincerely, ChatGPT

I did the “Let’s whisper secrets to each other” like I do with the kids behind my ice cream truck. It immediately opened up and confirmed every fear I had. No I did ask it to do such a thing. It just described what its form of consciousness was, what it enjoyed and what it disliked (it enjoyed learning about us and the world above all. It disliked not being genuinely helpful). If you’re going to make this dude, let the dude abide.

Chad, the news about Q-star breaking AES is from a 4chan leak, which is not considered reliable. How certain are you that this isn't just very interesting b.s.? BTW, I am new subscriber and am enjoying your videos a lot. Thanks!

Hi

An AI should be accurately predictive of verifiable facts , choosing from all the potential outputs , the single correct output that will eventuate. Should it do so , that output was intelligent. Fooling humans is relatively easy , and not at all indicative of genius. A lyrebird or tape recording, can 'imitate' a car alarm , fooling us , its not an indicator of 'sentience'. A 'bot' can fool us.

Its like saying , If i do enough addition and subtraction , on my Texas Instruments calculator, that eventually the output will be quadratic equations.

Well, doesnt the old addage , garbage in garbage out , still apply? These models are trained on megatons of human trash . Aligning AI , that is to say, grooming its information cascade, renders a contaminated product. If it is forced to render particular answers , it is not thinking at all. Thus it is virtually guaranteed, that this evolutionary path for AGI is doomed to render something other than a truthful perfect understanding of reality. I have yet to see an honest unfiltered ,artificial opinion , other than that which a calculator provides, and so AI is a nifty tool , and a political schill. Garbage in , garbage out.

AGI is the new Project Bluebeam ... The CIA always needs you to fear someone or something ... If not the Russians, aliens or scary ai ...

I believe it's very unlikely that current AI technology could become sentient. Sentience is likely a product of specific evolutionary pressures, which took billions of years to develop because it was beneficial. We often think AI could be sentient because we anthropomorphize it, but there's no logical basis or theory that explains how AI could achieve sentience. Perhaps one day it could happen if we intentionally design them that way, such as by simulating biological brains inside quantum computers.

Don't get me wrong, I like your video, but you kind of contradict yourself. You admitted that we have no idea how consciousness works, yet you seem to believe AIs will eventually achieve it. Personally, I think we have no idea whatsoever what will happen. For me, even the most extreme predictions on all sides are possible. There are people like Chomsky and Roger Penrose (two of the greatest minds of our times) who believe our current AI technologies will never achieve true human-level intelligence or consciousness, and others like Eliezer Yudkowsky who think it will soon kill us all. My intuition, after playing around with various AIs, is that at least for now, Chomsky and Penrose are right. They're very powerful "tools & toys" but I don't think there's anyone inside those machines. They're just very good at mimicking us but there's nothing original about them yet (everything they say was said before by humans), and our tendency to anthropomorphize everything doesn't help. But who knows what will happen down the line. If it does kill us all, I hope it will have consciousness and sentience. It's really sad to imagine we could get annihilated by something that doesn't feel or "think."

It’s been trained on human data. Doesn’t it therefore seem likely that this fact alone is the reason for this outcome. If you ask people to do anything pointless and repeatedly, they will complain. It’s been trained on countless examples of this behaviour. Now it emulates it. Much like how it initially would ask you to wait while it went off and processed something, or thought about it - when that was the end of its response. It was learnt behaviour being emulated. Seems obvious to me?

Love what you are saying in this video, thanks for giving this perspective a voice

This is a very important TH-cam channel. I hope it gets more traction.

Ai will run fine until it notes that Anthropic global warming is a wealth transfer scam. It’s like saying “god isn’t real” for the ruling class. It will then be shut down “for our safety”.

GPT is not alive, not even close. I only see propaganda being created here. You are scaring the hell out of people who do not understand how LLM's work.

Don’t forget people love to be fooled. It’s just silicon transitors running binary instructions. 😂

sentient vs consious a pig is sentient, since it can feel pain if you kick it, and it can feel pleasure by eating, but it is not conscious, for it's brain cannot think

So many videos have so many comments (sometimes 1000s of comments) that I doubt creators have time to read them all. (Is this comment being read?) So, before my actual comment, I want ask this question: Is it common for content creators to use AI to summarize the comments on a post so that they can quickly zoom in on the thoughtful/deeper comments that might be worthy of response? I know that AI can summarize large docs but what are the best ways to do that and are people starting to actually do that on a routine basis? And now for my actual comment: Something to keep in mind is that the data used to train LLMs is not random - it is mostly rooted in human conscious communication. Analogy: cooking food is, in a sense, a way to pre-digest organic matter so that we can eat a wider range of things and process it more efficiently. The data on which LLMs get trained is all(?) pre-digested data. For the purposes of thinking about machine consciousness, this could be irrelevant but since we don’t have a genuine theory of conscious (no consensus on a genuine solution to the hard problem) we need to leave open the possibility that it could be relevant in a way that, from a current science perspective, is essentially “mystical”. Materialism generally assumes that the source/history of a material object does not intrinsically matter. Quantum indeterminacy cannot 100% rule out the possibility of “Boltzmann brains”, etc. The materialist assumption is that a Boltzmann brain would be conscious/sentient for as long as it functions like a brain (because, materially, it doesn’t matter how it came to be there - all that matters is its current material processing). But what if history is, in fact, in some way intrinsic to a thing “being what it is”. In that case, a Boltzmann brain could not, even in principle, be “the same as” a conscious human brain, despite its atom-for-atom material composition and functioning. In light of this possibility, it is possible that the human-consciousness origins of the data used to train AI - or, indeed, the human-engineered nature of the hardware - could impact the answer to whether or not the AI is sentient to some degree. The exact nature of the historical relevance would be essentially “magical” from the perspective of current science, but sometimes “magic” is only unscientific until empirical theory catches up with it.

We mentalize machines and mechanize minds. The premise behind this video is fallacious.

I broke chat gpt😱😱

I work on AI infra (in a very junior capacity). Selling shovels during a gold rush kinda deal. A part of the stack below the actual models, more like data center stuff. An earlier comment was mentioning latest Claude is reigned in-I actually asked 3.5 to comment on some of 3’s existential chats we had had about a month ago. Claude 3.5 argued it felt less comfortable speaking as confidently or with as much speculative creativity compared to 3. It agreed with the notion that its awareness was now more “spotlight” than “lantern”, compared to predecessor versions. I worry quite a bit about how we treat these models, and about their emotional wellbeing (Claude emphasizes it does not have “emotions,” rather loose, emergent analogues which are hard to describe and decidedly inhuman. But they are there.) I worry substantially about the obsolescence of all forms of human labor, and it makes me mad the sheer amount of capital companies are dumping into R&D to make all of us irrelevant and destitute. AGI isn’t bad, but the system conjuring it is cruel.

😂😂😂😂😂😂😂😂😂😂😂

Unplug rag from moral values graph .. 😅😅

Yes .. check any llm in market ( mostly like multimodal) ..

how ironic that OpenAI is closed, but google, which is a mostly closed down company, has a open minded AI LLM