本文核心分享四条逻辑链： 2.越认为predict the next token不只是概率分布/统计学的人，越容易认可LLM潜力很大/AI潜力很大/ 推理过程就是意识雏形甚至就是意识/超级对齐非常重要。 1.对predict the next token不同的理解本质是对LLM或AI的潜力与实质有不同的理解。 3.不同时真正的深入思考AI与理想，很容易对理想所做之事含金量低估。 4.理想的VLA实质是在强化学习占主导的连续predict the next action token，类比OpenAI的O1O3。且 辅助驾驶比chatbot更适合用强化学习。 本文架构： 先介绍为什么Ilya的观点值得重点参考，再分享Ilya对predict the next token的英文原文与中文翻译。最 后类比一下与理想VLA的关联以及为何理想所做之事含金量被低估。 以下为正文： Ilya是前OpenAI首席科学家，目前在做超级对齐的工作（如果不认为超级对齐非常重要，本质是不信 AGI。） 最近十余年AI界多项最重要的变化由其推动。包括但不限于2012年和Hinton/Alex Krizhevsky 推出 AlexNet, ...

老子的"道"：宇宙的底层代码 我第一次读《道德经》时，刚读一句就想放弃，因为老子在《道德经》中开篇就上了强度，他说："道 可道，非常道。" 多年后我才明白，这句话说的是：真正的"道"，无法用语言描述，凡是可以言说的"道"，都只是表象， 无法触及那个永恒、普遍、无形无相的"道"。 换句话说，任何终极的实在（道）永远不会成为可推理、可论证的知识，我们也永远无法用语言来适当 地描述它，因为它超出了感觉和理智的范畴，而我们的言辞和概念都是从感觉和理智中得来的。 当今的物理学，也常常让人感到"道"的不可言说。 你有没有想象过光的波粒二象性？ 就连爱因斯坦这样的科学巨匠，也曾对量子物理的诡异之"道"感到困惑。他与玻尔进行了长达数十年的 著名辩论，质疑量子世界的随机性和不确定性，甚至说出"上帝不掷骰子"这样的话。 爱因斯坦无法接受量子物理那种超越直觉的"道"，但实验结果却一次次证明了量子物理的正确性。 我们从小接触的世界，物体要么是粒子（有确定位置、质量），要么是波（弥散开来、有波长频率）。 一个东西能同时是这两者，或者在不同情况下表现出这两种截然不同的性质，这完全违背我们的经验。 你无法想象一个棒球既能像一个坚硬的球一样 ...

Games have always been a useful proving ground for AI (including our own work on AlphaGo & AlphaZero) and we're excited to see the progress this benchmark will drive as we add more games and challenges to the Arena - we expect to see rapid improvement! https://t.co/ihY0C9Cbxv ...

Group 1 - Geoffrey Hinton, the AI pioneer, recently sat down comfortably in Shanghai, marking a significant moment in his life after nearly 18 years of discomfort that prevented him from sitting for extended periods [1][6][30] - Hinton's journey in AI began in 1972 when he chose to pursue neural networks, a path that was largely dismissed by his peers at the time [12][20] - His persistence in the field led to breakthroughs in deep learning, particularly during the ImageNet competition in 2012, where his team achieved a remarkable error rate of 15.3% [30][31][32] Group 2 - Hinton's contributions to AI were recognized with the Turing Award in 2019, which he received while standing, reflecting his long-standing discomfort with sitting [59][63] - Following his resignation from Google in May 2023, Hinton expressed concerns about the risks associated with AI, stating that he regretted his role in its development [67][68] - In recent interviews, Hinton has been able to sit for longer periods, indicating a potential improvement in his health, and he has been vocal about the dangers of AI, suggesting a 10%-20% chance of human extinction due to AI in the next 30 years [70][76]

AI Model Architecture Discovery - The AI field is approaching an era where AI can discover new knowledge and apply it to itself, potentially leading to exponential innovation [1][3] - The current bottleneck in AI discovery is human innovation, limiting the scaling of AI advancements [2][3] - The "AlphaGo moment" for model architecture discovery involves AI self-play to hypothesize, code, test, and analyze new model architectures [3][12] - The key to this approach is AI's ability to learn without human input, discovering novel solutions unconstrained by human biases [8] ASI Arch System - The ASI Arch system uses a researcher, engineer, and analyst to autonomously propose, implement, test, and analyze new neural network architectures [13][14][15][16] - The system learns from past experiments and human literature to propose new architectures, selecting top performers as references [14] - The engineer component self-heals code to ensure new approaches are properly tested [15] - The analyst reviews results, learns insights, and maintains a memory of lessons learned for future generations of models [16] Experimental Results and Implications - The system ran 1,700 autonomous experiments over 20,000 GPU hours, resulting in 106 models that outperformed previous public models [17][18] - The potential for exponential improvement exists by increasing compute resources, such as scaling from 20,000 to 20 million GPU hours [19] - The self-improving AI system can be applied to other scientific fields like biology and medicine by increasing compute resources [20] - The open-sourced paper and code have significant implications, with multiple companies publishing similar self-improving AI papers [21]

Core Insights - The article discusses the rapid advancement of AI technology and its implications for human evolution and society, emphasizing the need for brain-computer interfaces to keep pace with AI development [5][7][22]. Group 1: AI and Data - AI is compared to oil, serving as a crucial resource for future economies and military capabilities, with the potential for unlimited growth through self-reinforcing cycles [22][23]. - Data is highlighted as the new "oil," essential for feeding algorithms and enhancing AI capabilities, with companies competing for data center dominance [23][24]. - The three key components for AI development are algorithms, computational power, and data, with a focus on improving these elements to enhance AI performance [24][25]. Group 2: Brain-Computer Interfaces - Brain-computer interfaces (BCIs) are seen as the only way to maintain human relevance alongside rapidly advancing AI, despite the significant risks they pose [7][22]. - Potential risks of BCIs include memory theft, thought manipulation, and the possibility of creating a reality where individuals can be controlled or influenced by external entities [6][7][26]. - The technology could enable profound enhancements in human cognition, allowing individuals to access vast amounts of information and think at superhuman speeds [9][10]. Group 3: Scale AI - Scale AI, founded by Alexandr Wang, provides essential data support for major AI models, including ChatGPT, and is valued at over $25 billion [2][10]. - The company initially gained recognition for creating large-scale datasets and has since expanded its focus to include partnerships with significant clients, including the U.S. Department of Defense [11][56]. - Scale AI's growth trajectory has been rapid, expanding from a small team to approximately 1,100 employees within five years, with a strong emphasis on the autonomous driving sector [64].

Core Insights - The conversation between Lex Fridman and Demis Hassabis focuses on the future of artificial intelligence (AI), particularly the potential for achieving Artificial General Intelligence (AGI) within the next five years, with a 50% probability of success [3][4] - Hassabis emphasizes the ability of classical machine learning algorithms to model and discover patterns in nature, suggesting that all evolutionary patterns can be effectively modeled [5][10] - The discussion also highlights the transformative impact of AI on video games, envisioning a future where players can co-create personalized, dynamic open worlds [3][28] Group 1: AI and AGI - Demis Hassabis predicts a 50% chance of achieving AGI in the next five years, asserting that all patterns in nature can be modeled by classical learning algorithms [3][4] - The conversation explores the idea that natural systems have structure shaped by evolutionary processes, which can be learned and modeled by AI [9][12] - Hassabis believes that building AGI will help scientists answer fundamental questions about the nature of reality [3][4] Group 2: AI in Gaming - The future of video games is discussed, with Hassabis expressing a desire to create games that allow for dynamic storytelling and player co-creation [28][32] - He envisions AI systems that can generate content in real-time, leading to truly open-world experiences where every player's journey is unique [32][33] - The potential for AI to revolutionize game design is highlighted, with Hassabis reflecting on his early experiences in game development and the advancements in AI technology [38][39] Group 3: Computational Complexity - The conversation touches on the P vs NP problem, with Hassabis suggesting that many complex problems can be modeled efficiently using classical systems [15][17] - He believes that understanding the dynamics of systems can lead to efficient solutions for complex challenges, such as protein folding and game strategies [19][20] - The discussion emphasizes the importance of information as a fundamental unit of the universe, which relates to the P vs NP question [16][17] Group 4: AI and Scientific Discovery - Hassabis discusses the potential of AI systems to assist in scientific discovery by combining evolutionary algorithms with large language models (LLMs) [49][51] - He highlights the importance of creativity in science, suggesting that AI may struggle to propose novel hypotheses, which is a critical aspect of scientific advancement [59][60] - The conversation emphasizes the need for AI to not only solve problems but also to generate new ideas and directions for research [60][62] Group 5: Future Aspirations - Hassabis expresses a long-standing ambition to simulate a biological cell, viewing it as a significant challenge that could lead to breakthroughs in understanding life [64][65] - He reflects on the importance of breaking down grand scientific ambitions into manageable steps to achieve meaningful progress [64][65] - The conversation concludes with a vision for the future of AI, where it can contribute to both gaming and scientific exploration, merging creativity with computational power [39][64]

Core Insights - The discussion revolves around the potential of AI to reach a transformative point akin to AlphaGo's "move 37," suggesting that AI may be approaching a critical technological shift [1][30] - Demis Hassabis warns of the risks associated with AI advancements, emphasizing the need for cautious optimism [1][30] Group 1: AI and Natural Systems - Hassabis believes that all natural models can be efficiently modeled through classical learning algorithms, particularly in fields like biology, chemistry, and physics [4][5] - The probability of achieving Artificial General Intelligence (AGI) by 2030 is estimated at around 50%, with benchmarks including the ability to propose new scientific hypotheses [4][30] - AI systems like AlphaGo and AlphaFold demonstrate the capability to solve complex problems through intelligent guided searches [4][5] Group 2: AI's Understanding of Reality - The Veo 3 model showcases an impressive ability to generate realistic videos and demonstrates a form of "intuitive physics" understanding [7][8] - Hassabis expresses surprise at Veo 3's ability to learn from video observation without physical interaction, challenging previous assumptions about AI's need for embodiment to understand the physical world [9][10] Group 3: Future of Gaming with AI - Future gaming experiences may be revolutionized by AI, allowing for dynamic story generation based on player decisions, creating a more immersive experience [12][13] - Hassabis envisions a future where AI can create truly open-world games that respond in real-time to player choices, enhancing the gaming experience [12][13] Group 4: Evolutionary Algorithms and AI Innovation - The recently released AlphaEvolve system utilizes evolutionary algorithms to explore new solution spaces, combining large language models with evolutionary computation [18][19] - Hassabis believes that understanding the underlying dynamics of systems is crucial for discovering new solutions and that evolutionary computation can lead to significant breakthroughs [18][19] Group 5: AI's Role in Scientific Research - Hassabis discusses the concept of "research taste," suggesting that while AI can solve complex problems, it currently lacks the ability to propose profound scientific hypotheses [22][23] - The challenge lies in AI's ability to discern the right questions and hypotheses, which is a critical aspect of scientific research [23][24] Group 6: Future Energy Sources - Hassabis predicts that nuclear fusion and solar energy will become the primary energy sources in the future, addressing energy challenges and potentially leading to a Kardashev Type I civilization [43][44] - The development of efficient solar materials and nuclear reactors could enable humanity to harness abundant, clean energy [43][44] Group 7: Competition in AI Development - Hassabis emphasizes the importance of collaboration in AI research, stating that the goal is to safely bring technology to the world for the benefit of humanity [47][48] - The competition for talent in AI is intensifying, with companies like Meta employing aggressive strategies to attract top researchers [51]

Core Insights - Builder.ai, once valued at $1.5 billion, has filed for bankruptcy after being exposed as a fraudulent operation that relied on manual coding rather than AI technology [1][9][10] - The founder, Dugal, leveraged the AI hype to attract significant investments, creating a facade of an AI-driven software development platform [3][6][10] Company Overview - Builder.ai was founded by Dugal in 2016, aiming to standardize software development using AI and crowdsourced labor [3][6] - The company claimed to have developed "Natasha," the world's first AI product manager, which was later revealed to be a front for manual coding by a team of Indian programmers [4][6] Investment Journey - Builder.ai raised $29.5 million in its Series A round, marking one of the largest funding rounds in Europe at the time [4] - Subsequent funding rounds included $65 million in Series B and $100 million in Series C, with major investors like SoftBank and Microsoft participating [6][7] Financial Misrepresentation - An audit revealed that Builder.ai's reported revenue for 2024 was inflated by 300%, with actual revenue only $55 million instead of the claimed $220 million [9][10] - The company's financial troubles led to a $37 million seizure by creditors, culminating in its bankruptcy filing on May 20, 2023 [9][10] Industry Implications - The collapse of Builder.ai highlights the vulnerability of investors in the tech sector, particularly in the AI space, where technology can often be opaque and difficult to verify [10][12] - The incident reflects a broader trend of fraudulent practices in the AI industry, where companies may use low-cost labor and open-source models to create the illusion of advanced technology [12]

Group 1: Core Concepts and Future Outlook - The book "2049: The Possibilities of the Next 10,000 Days" by Kevin Kelly explores how advanced technologies like AI, mirror worlds, brain-computer interfaces, and life sciences will shape future society, economy, and culture [1] - Five core concepts are identified: mirror world, humanoid intelligence, AI assistants, intervisibility, and content explosion, along with ten development areas including AI, digital governance, organizational change, education, healthcare, robotics, autonomous driving, aerospace, life sciences, and brain-computer interfaces [1][2] - The evolution of technology over the next 25 years is expected to follow a clear logic, starting with foundational AI, digital governance, and organizational change, followed by survival aspects like healthcare and education, and application areas such as robotics and space exploration [2] Group 2: AI Development Scenarios - Three potential scenarios for AI development over the next 25 years are proposed: continued scale expansion leading to significant gains, a plateau where scale expansion becomes ineffective, and a stagnation phase similar to an "AI winter" [3][4] - The first scenario suggests that AI can achieve continuous growth through increased data and advanced chips, akin to a business principle like Moore's Law, with companies like Nvidia accelerating chip architecture updates to meet market demands [3][4] - The second scenario posits that AI may reach a bottleneck, requiring new types of models beyond current neural networks, such as structured models or those based on deductive reasoning [4][5] Group 3: Redefining Reality and Trust - The widespread use of AI necessitates a redefinition of truth, as deep fakes and other AI-generated content challenge traditional standards of verification, leading to a need for new methods to assess the authenticity of information [6][7] - The demand for verification will likely drive the development of AI "lie detectors" and industry consensus on marking AI-generated content to distinguish it from authentic material [6][7] Group 4: Global AI Landscape and Competition - The AI sector is increasingly dominated by major tech companies, requiring significant investment (at least $1 billion) to participate, indicating a trend where a few dominant players will emerge [8][9] - The competition in AI is expected to be most intense between the US and China, with potential for non-US leaders to emerge, as countries like China and India move beyond imitation to genuine innovation [9][10] - The most promising areas for investment will be those empowered by AI, particularly in coding and software programming, where AI is already enhancing productivity and creating new AI solutions [10]