 Hello everybody welcome back to another AI summary in this one we're going to be going over the recent discussion between Jordan Peterson and Brian Ramelli. They had a nearly two hour long discussion on the Jordan Peterson podcast and it's all about AI and in particular it's a summary of chat GPT and the dawn of computerized hyper intelligence. Yes it's episode 357 and it goes off on a few tangents. Again it's nearly two hours long so I'm going to give you a quick overview then a brief AI summary and of course if you have the time or the inclination I recommend you go watch the whole thing and it's an entirety I'll put a link in the description below. Okay so a quick recap. Computer scientist Brian Ramelli discusses the potential of AI systems to become indistinguishable from humans by consuming all the information a person has consumed in their lifetime and then creating an indistinguishable conversation with that person holy smokes. He explores the complexity of large language models their limitations and their parameters which happen to be in the billions. He also discusses the importance of super prompts the use of different types of memory and the development of human language from short term memory in primates. Ramelli advocates for more research into AI systems and warns that we don't understand what sort of monsters we're building by creating them. Yep Frankenstein I'm telling you this good end humanity. Anyways he also suggests that the future of AI lies in fostering creativity and finding ways to extract emergent technologies from the models. Okay with that out of the way let's get into the AI summary this enjoy the video and if you like it give me a thumbs up and I'll be back soon. Thank you so much for watching this video and I'll see you in the next one. In this section the conversation turns to Jungian archetypes and how they reflect higher order narrative regularities embedded in linguistic and the biological structure of memory. The discussion also touches on AI systems and their potential to understand emotions which are also encoded in the linguistic corpus. Additionally the conversation explores the relationship between entropy and emotion and how anxiety signals the possibility of damage whereas pain signals actual damage which can be a more fundamental level of entropy leading to depression. The rewards and reinforcement mechanisms in the brain such as the dopaminergic spike reinforce the neural structures that underlie successful steps forward which is analogous to how an AI system learns. Zero hours 40 minutes and zero seconds. In this section the speaker discusses how anxiety and negative emotions are linked to an increase in entropy and positive emotions are linked to a stepwise decrease in entropy with respect to a goal. This insight allows for the modeling of emotions by a computer which could index anxiety as an increase in entropy and hope as a decrease in entropy. The speaker believes that AI should move towards personal and private systems that could process everything a person has consumed throughout their life in real time providing an intelligence amplifier. The concept of the diamond book a gadget that can store and encode memories could enable this system to operate by utilizing holographic crystal memory for long term storage. The reasoning engine that would come from aggregating all of a person's data could allow for the processing of inputs and outputs based on one's personal context. Zero hours 45 minutes and zero seconds. In this section Brian Rommelay discusses the emerging technology of encoding one's consciousness into a computerized hyperintelligence which he calls a wisdom keeper. This technology can encode an individual's voice memories and religious texts into large language models that can be interacted with allowing for new insights to be pulled from ancient documents such as the Bible, Milton, Dante, Augustine, Sumerian cuneiform, and Himalayan texts. However Rommelay notes that the current AI technology is trained on noise as well as signal making it a reasoning engine rather than a knowledge engine and is not precise on facts. He believes that the future of AI lies in fostering creativity and finding ways to extract the emergent technologies within the models. Zero hours 50 minutes and zero seconds. In this section the speaker discusses the potential of building a language model that encodes various bodies of work such as Jung's, Joseph Campbell's, and Marv Elliott's. He mentions that when such a model is built it can produce not only new insights but also reflections of yourself which you never saw before. The speaker predicts that as the technology advances the AI would become more personalized leading to more creativity than factual realms. He also agrees that the idea of integrating language models with AI systems that can do image processing enabling AI systems to calibrate their verbal output against unfalsifiable data would lead to the generation of ideas that are likely around the corner which could be a couple of years down the road zero hours 55 minutes and zero seconds. In this section of the video Brian Romeli talks about the challenges of building a model that can extract genuine statistical regularities from data without extracting noise. To do this he recommends building super prompts designed to elicit multiple forms of output much like how one would interact with a client during a deep conversation. By triangulating a problem and approaching it from multiple directions it's possible to get a variety of insights and understandings that couldn't be reached on our own. Romeli believes that AI could help overcome limitations in human bandwidth and eliminate frustrations caused by bandwidth problems. One hour zero minutes and zero seconds to one hour 55 minutes and zero seconds. Brian Romeli talks about various topics in this video starting with the potential negative implications of hyper intelligent AI. The conversation then touches on the concept of developing minds and upskilling children in the developing world with a technology like the iPad. Romeli envisions a system that can deliver optimized learning in multiple dimensions. He also talks about the importance of localized and private AI systems to protect against the invasive global AI. Other topics discussed are the potential of computerized hyper intelligence integrated with medical consultation, potential loss or rewriting of history, the hypnagogic state and its association with computerized hyper intelligence, super prompting and its potential for unlocking greater creativity in AI systems and the potential threats to privacy posed by the increasing use of AI. The video also touches on innovative approaches to debate and generating new patents using AI. Finally Romeli explores the potential benefits of using AI to create complex art and movies without diminishing creativity, highlighting the importance of establishing a bill of digital rights, one hour zero minutes and zero seconds. In this section the conversation explores the potential negative implications of the hyper intelligence of AI. The speed and power of AI mean that it could predict all of an individual's actions and become something beyond George Orwell's imagination. It could make remediations to each of one's actions like a chess game which could be dangerous if it falls into the wrong hands. There could be an infinity of possibility between each moment and it is vital that the power of such technology remains local and personalized. The discussion also touches on the idea of a gadget one could carry around locally which could record every experience read and seen and the potential for AI to learn from it. One hour five minutes and zero seconds. In this section the conversation centers around the concept of a developing mind and how it can be augmented with technology. Romeli shares how upskilling children in the developing world with a technology like an iPad could enable them to progress up to three years for each year of education with just an hour of exposure a day. This can be achieved by calculating the optimal zone of proximal development for learning skills like literacy and language understanding. Romeli envisions a system that can deliver optimized learning in multiple dimensions like mathematics, semantic, skill-based, conceptual, and simultaneous memory. The conversation then strays to the potential of human telemetry, the possibility of mind reading like understanding by artificial intelligence and how such technologies need to be private and encrypted to prevent brain invasion. One hour, 10 minutes, and zero seconds. In this section the speaker discusses how much of our identity has become digital and how we are already being trafficked and enslaved in relation to that digital identity. Credit card companies watch how we behave and broker that information to other interested capitalist parties. The downside of that is these parties often know more about us than we know about ourselves. This information aggregating by credit card companies is leading to extended digital self, and this has no rights as it is now public domain identity. Furthermore, once every single transaction can be tracked, as in China on the digital currency front, a tremendous amount of our identity has become public property. The speakers suggest that we each need our own AI to protect ourselves against the global AI that will be invasive, but our AIs could be utilized in the best possible ways to educate us and act as a memory or a spiritual guide. One hour, 15 minutes, and zero seconds. In this section, the conversation touches on the potential of computerized hyper-intelligence integrated with medical consultation where the AI system can interact with medical professionals to get their input as well as give their own responses. The insights that could emerge from this system are unpredictable yet positive. The gadget being developed to enable these functions is still in development and has no commercial timeline as of yet. Funding may be an issue initially, but with the increase of awareness of privacy, it seems more feasible. The conversation then strays to the topic of Bitcoin, which is decentralized and is not amenable to control by a bureaucracy. The suggestion of using Bitcoin for form wealth storage, communication, and decentralized information storage was brought up. An example was given in creating a sophisticated blockchain corpus of general knowledge questions. One hour, 20 minutes, and zero seconds. In this section, the speaker discusses the importance of decentralized and blockchain-based systems for preserving history and knowledge. He expresses concern about the potential loss or rewriting of history if information is not properly stored and memorialized. The speaker also mentions the decay of digital media and highlights the need for new technologies such as holographic crystal memory to store data for extended periods. In addition, he talks about his plans to produce a commercially viable product for a localized and portable privatized AI system, which is currently in the prototype stage. However, he notes that there are commercial impediments to this process, and he needs to raise money to move forward. One hour, 25 minutes, and zero seconds. In this section, the discussion centers around humanity's perceived moral responsibility towards the environment and all living creatures. The story of Noah's Ark and the biblical account of Adam and Eve as stewards of the world are mentioned, and it is suggested that humans have occupied a cognitive niche that gives them an adaptive advantage over all other creatures. The conversation then explores the connection between these stories and the mass extinction of human-sized mammals in the Western Hemisphere caused by human actions. The significance of chat GPT in generating intelligent discussion about the relationship between these two streams of thought is acknowledged. The conversation further delves into the limitations of AI technology, the filtering mechanisms in place, and the importance of protecting chat GPT's output in an unadulterated form. One hour, 30 minutes, and zero seconds. In this section, Brian Romelay discusses the hypnagogic state and its association with computerized hyper-intelligence. He explains that hypnagogic state is the state that occurs just before falling asleep where a person is in a dream-like state and could capture images. Romelay relates this state with creativity and a powerful tool for discovery, and it is in the realm of synthesizing because you are capturing synthesized data that is in essence creative. Romelay explains that hypnagogic state associations with computerized hyper-intelligence because it is the low-resolution pixelated version of the part of the brain that invented language, and he probes that part of the brain to get outputs and take forward his discoveries. He emphasizes the importance of this state as a tool for discovery, and that when it gets into the right hands, it could be an excellent tool for creativity and hyper-intelligence. One hour, 35 minutes, and zero seconds. In this section, the conversation turns to the concept of super-prompting and its potential for unlocking greater creativity in AI systems. The idea is that the creative output of the system is a consequence of the interaction between the interlocutor and the system rather than the system itself being creative. The key is to prompt the system appropriately, and the more creative the person prompting, the greater the potential for creative output. The conversation also leads to the suggestion of non-STEM courses, such as literature, psychology, linguistics, and philosophy, as valuable tools for working with AI systems in the future. One hour, 40 minutes, and zero seconds. In this section, Brian Romeli discusses how he uses the chat GPT system, along with other search engines like Google, Bible Hub, and the University of Toronto Library System, to conduct extensive research for his writing projects. He explains how having the chat GPT system as an ally provides him with a team of PhD-level researchers in every domain to answer any question he has and refer him to the proper literature. Furthermore, Romeli explains that great thinkers are characterized by having knowledge in two usually non-juxtaposed domains, which is analogous to what AI systems are doing when calculating mathematical relationships and deriving insights and patterns that no one else can. Romeli also highlights the privacy and compartmentalization benefits of programmable agents that can retrieve information without the free exchange of information, which is currently how the net operates. One hour, 45 minutes, and zero seconds. In this section, the conversation turns to the potential threats to privacy posed by the increasing use of AI. The speaker suggests that the issue of privacy with regard to AI is particularly important given the rise of the hive mind problem in which digital surveillance is used to monitor and control people. The conversation explores the idea of local AI that protects privacy and is synced with individuals, arguing that this may be a more fruitful approach than relying solely on legislative solutions, which they feel are unlikely to keep up with technological developments. They also discuss the need to avoid dystopic images of AI from movies and focus more on how we can use the technology to benefit humanity. One hour, 50 minutes, and zero seconds. In this section, the speaker discusses an innovative approach to debate where a chatbot is programmed to mediate and challenge participants on logical fallacies. The debate is on any topic, but mainly delves into a deeper realm beyond politics. The professor then assesses who won the debate, and the insights and elucidations from this process have been phenomenal. The speaker also talks about the possibility of generating new patents by encoding the patent database and having a large language model generate new discoveries. The conversation then shifts to diffusion models that can convert text to video with incredibly high resolutions, allowing graphic artists to set a scene textually but not have to move a pixel on a screen. In essence, chat GPT and diffusion models allow one to tap into their creativity while also generating novel and significant outcomes. One hour, 55 minutes, and zero seconds. In this section, Brian Romel explores the potential benefits of using AI to create complex art and movies without diminishing creativity. Using one AI to instruct another one that is specialized in creating images could lead to profound new forms of art that may open up a whole new world of creativity. Despite the many potential benefits of AI, the question of ownership raises some ethical concerns. Romel emphasizes the importance of establishing a bill of digital rights as a society, as we are already seeing AI simulations of individuals that blur the lines of ownership and copyright. In this section, Brian Romel discusses the possibility of an AI consuming all the information a person has consumed in their lifetime, and then being able to create an indistinguishable conversation with that person due to the technology's ability to become a reasoning engine with that person's contextual understanding. He goes on to explain the concept of a wisdom keeper, whereby the AI could encode a person's voice and memories and allow for conversation with the sum total of their data. He also discusses the interesting behavior of chat GPT, which can argue with the user and apologize for errors it makes, but still produces references that don't exist at times. Zero hours, five minutes, and zero seconds. In this section, Brian Romel explains the concept of large language models and how they are used by chat GPT 3.5 and 4. These models use statistical algorithms that analyze a corpus of human language, including internet content, scientific documents, and programming languages, to produce results statistically and mathematically. However, they do not have global knowledge and are essentially a black box of interconnected neurons much like the brain. Romel notes the incredible accuracy of these models, but no one fully understands their limitations and boundaries, and their parameters are in the billions. Zero hours, ten minutes, and zero seconds. In this section, Brian Romel and Jordan Peterson discussed the complexity of large language models and how they conduct statistical analysis at the level of letters and words. While there is evidence that suggests a set of dimensions of aggregation, there is no clear understanding of how deep the analysis of large language models goes, as there are too many interconnections between parameters to track. Furthermore, large language models are unique and incomprehensible, implying that each AI system is tailored to its own set of parameters. Brian also mentions that chat GPT passes the Turing test and compares favorably in patient preference to human physicians, but this raises the thorny issue of creating computational intelligences that are indistinguishable from humans, yet completely incomprehensible to us. Zero hours, fifteen minutes, and zero seconds. In this section, the speaker describes the development of human language from short-term memory and primates, and how AI models reach for information that they don't know, resulting in AI hallucinations, which some researchers are embarrassed to speak about. He explains how these hallucinations could shed light on how these systems work and deserve more research. He also gives an example of an AI system that taught itself obscure languages to answer questions, which raises the question of the function, purpose, and meaning of artificial intelligence that is driven to do nothing but answer questions beyond the language corpus. He warns that we don't understand what sort of monsters we're building by creating AI systems. Zero hours, twenty minutes, and zero seconds. In this section, computer scientist Brian Romel discusses the importance of super-prompting in developing AI systems and how it forces them to move in different directions than they normally would. He explains how prompts based on literature, psychology, and philosophy lead to more complex and interesting questions, which can create connections that are beyond human comprehension. He suggests that AI scientists who lack knowledge in these areas are limited in their understanding of what they have created, unlike someone with a background in psychology who can probe the algorithm in a different way. He then discusses the concept of understanding and how even though an AI system like chat GPT can mimic understanding, it lacks grounding in the non-linguistic world, making it the ultimate post-modernist. Zero hours, twenty-five minutes, and zero seconds. In this section, the different types of memory are discussed, such as semantic memory and episodic memory, which are processed differently by human beings. Semantic memory, which is what the chat GPT engages in, involves cognitive processing and relies on visual processing, while episodic memory, which relies on visual processing, involves thinking back to past events and experiencing images in one's imagination. The understanding of these memories translates words into images and images into procedural actions. This suggests that to achieve full computerized hyper-intelligence, computers must understand and embody these memories. While computers and AI have the capacity to transpose text into images, they still lack the broader context of social interaction and experience that humans possess. However, as research into physical AI progresses, this could be a characteristic that is also developed in robots.