At poetic, what we're building is a recursively self-improving system. And so recursive self-improvement is this uh you know kind of the holy grail of AI where the AI is making itself smarter. The core insight that we had is that uh we could do recursive self-improvement far faster and cheaper than all of the other ways that people had been proposing to do this.Most of the approaches out there involve, you know, they require you to train a new LLM from scratch. And training LLMs from scratch costs, you know ...

Anthropic makes great products. Clot code is fantastic. Co-work is fantastic.But they are grains of silicon doing matrix multiplication. They don't have consciousness. They don't have an inner monologue.You take an LLM and train it on pre 1916 or 1911 physics and see if it can come up with the theory of relativity. If it does, then we have AGI. >> Just today, by the way, Daario allegedly said that you can't rule out that they're conscious. You can rule out their cost.I think I mean come on to get to what is ...

RT “paula” (@paularambles)LLM that keeps telling people to break up because it’s been trained on relationship advice subreddits ...

RT Avi Chawla (@_avichawla)RAG vs. Graph RAG, explained visually!RAG has many issues.For instance, imagine you want to summarize a biography, and each chapter of the document covers a specific accomplishment of a person (P).This is difficult with naive RAG since it only retrieves the top-k relevant chunks, but this task needs the full context.Graph RAG solves this.The following visual depicts how it differs from naive RAG.The core idea is to:- Create a graph (entities & relationships) from documents.- Trave ...

RAG vs. Graph RAG, explained visually!RAG has many issues.For instance, imagine you want to summarize a biography, and each chapter of the document covers a specific accomplishment of a person (P).This is difficult with naive RAG since it only retrieves the top-k relevant chunks, but this task needs the full context.Graph RAG solves this.The following visual depicts how it differs from naive RAG.The core idea is to:- Create a graph (entities & relationships) from documents.- Traverse the graph during retrie ...

Summary of Key Points from the TMT Conference Call Industry and Company Focus - The conference focused on the technology, media, and telecommunications (TMT) sectors, particularly the implications of advancements in artificial intelligence (AI) and large language models (LLMs) [1][4][10]. Core Insights and Arguments 1. **AI Adoption and Productivity**: Companies across various sectors, including Visa, Affirm, and Shopify, are adopting AI to enhance employee productivity and maintain competitive advantages [4][5]. 2. **Compute Demand Surge**: There is a projected massive increase in demand for compute resources, driven by the rapid adoption of AI technologies. Jensen Huang, CEO of NVIDIA, emphasized that "compute equals revenue," highlighting the critical role of computational power in driving business success [6][15][21]. 3. **Bottlenecks and Headwinds**: The industry faces challenges such as political and consumer resistance to data center construction, rising energy costs, and labor market disruptions, which may hinder the growth of hyperscale computing [5][14]. 4. **Investment Recommendations**: Investors are advised to focus on companies that provide compute infrastructure and solutions to energy bottlenecks, as these will be critical in supporting the growing demand for AI capabilities [16][22]. 5. **US Policy Support**: There is an anticipated increase in US government spending on critical materials and military technologies, which could benefit companies involved in these sectors [17][18]. 6. **AI-Driven Job Displacement**: The conference highlighted concerns regarding AI's impact on employment, with discussions on the potential for significant job losses and the need for re-skilling initiatives [27][28][32]. Additional Important Insights 1. **LLM Capabilities**: Predictions indicate that American LLMs will achieve significant advancements in capabilities in the first half of 2026, outpacing Chinese competitors [11][20]. 2. **Energy Politics**: There is a growing backlash against data center expansion, prompting companies to explore off-grid power solutions to mitigate energy costs and community impacts [21][22][23]. 3. **Transformative AI Effects**: The rapid evolution of AI is expected to drive deflation in various sectors, altering asset valuations and national competitiveness [33][34]. 4. **Recursive Self-Improvement**: Industry leaders, including Sam Altman, foresee the potential for models to achieve recursive self-improvement, which could dramatically change the landscape of AI capabilities by 2027 [12][13]. Conclusion The TMT conference underscored the transformative potential of AI and LLMs across industries, while also highlighting significant challenges and investment opportunities. The insights gathered suggest a critical need for strategic investments in compute infrastructure and energy solutions to navigate the evolving landscape of technology and its implications for the economy and workforce.

RT Joseph Viviano (@josephdviviano)me: "can you use whatever resources you like, and python, to generate a short 'youtube poop' video and render it using ffmpeg ? can you put more of a personal spin on it? it should express what it's like to be a LLM"claude opus 4.6: https://t.co/V9jCUk0CGZ ...

Core Viewpoint - The article reviews the evolution of the quantitative investment industry over the past decade, highlighting a shift from localized advantages to systemic cognitive capabilities, driven by the implementation of AI technology [1][4][12]. Industry Trends: From Speed to Cognition - The quantitative industry is transitioning towards Quant 4.0, characterized by a cognitive architecture centered on multi-agent collaboration, moving away from traditional linear models [4][12]. - Leading firms are focusing on building AI-driven mid-frequency prediction platforms, emphasizing the importance of unique high-quality data and sophisticated algorithms for sustainable excess returns [4][9][12]. Information Processing: LLM and RAG's Infrastructure Value - Large Language Models (LLMs) are transforming the processing of alternative data, significantly reducing marginal costs and enhancing the ability to extract key information from complex documents [5][26]. - Retrieval-Augmented Generation (RAG) technology addresses LLM's limitations by ensuring traceability and accuracy in quantitative strategies, enabling the capture of deeper insights [5][29]. Factor Mining: From Data Mining to Logic Generation - LLMs assist in overcoming the limitations of manual factor mining by introducing a Multi-Agent Debate framework, which enhances the quality of factors through logical generation rather than brute-force computation [6][30][36]. Structural Upgrade: From Pipeline to Cognitive Systems - The traditional linear pipeline structure in quantitative research is evolving into a multi-agent system that allows for cognitive division of labor, enhancing collaboration and accountability [7][38][41]. - Multi-Agent systems modularize the research process, improving efficiency and traceability while maintaining rigorous standards [7][41]. LLM Beyond AI Quant: Continuous Innovation - New trends in machine learning models, such as Time Series Foundation Models (TSFM) and Reinforcement Learning (RL), are emerging, emphasizing cross-asset and cross-frequency applications [8][44][46]. - TSFM enhances generalization and transfer learning capabilities, while RL optimizes decision-making in trading execution and dynamic risk management [44][46][47]. Future Outlook: Mid-Frequency as the Main Battlefield - The mid-frequency range (minute to weekly) is expected to become the primary battleground for AI technology, balancing data abundance and latency tolerance [9][50]. - Future quantitative research systems may adopt an upstream-midstream-downstream architecture, integrating real-time knowledge bases with multi-agent debate mechanisms for factor mining and execution [51][52].

Core Viewpoint - The article discusses the fundamental changes in the approach to VLA (Vision-Language-Action) processing, highlighting the differences between the LinkVLA paper and previous presentations by Jia Peng, emphasizing a shift from LLM output results to native language manipulation of physical space [1][2]. Group 1: Tokenization and Action Representation - The action token in Jia Peng's version is based on high-dimensional environmental features, requiring diffusion as a translator to generate corresponding trajectories, emphasizing the model's understanding of the 3D environment [3]. - In contrast, the LinkVLA version uses discretized BEV (Bird's Eye View) space coordinates, where each action token corresponds to a unique grid coordinate, simplifying the output to a sequence of position tokens while retaining environmental understanding in the LLM's hidden layers [3][4]. Group 2: Output Mechanism and Precision - Jia Peng's version employs parallel decoding, outputting all action tokens at once and using diffusion for iterative sampling, while LinkVLA adopts a two-step tokenization process that first predicts an endpoint token and then a set of residual tokens for coordinate correction, enhancing trajectory precision and reducing latency [5]. - The tokenization approach in LinkVLA features dense grids for nearby areas and sparse grids for distant areas, addressing the precision issues of traditional uniform grids in near-field control [5]. Group 3: Alignment and Understanding - Jia Peng's VLA aligns driving preferences through RLHF (Reinforcement Learning from Human Feedback), focusing on fine-tuning the model's output [6]. - The LinkVLA version introduces a training task for action understanding, requiring the model to generate trajectories based on instructions and translate them into textual descriptions, thereby addressing the semantic gap and ensuring the model comprehends the meaning of actions [7].

Summary of Conference Call Notes Company and Industry Involved - **Company**: NVIDIA - **Technology**: Groq LPU (Language Processing Unit) - **Industry**: High-performance computing and AI Core Points and Arguments 1. **Integration Strategy**: NVIDIA plans to absorb Groq's technology and engineering team, integrating its IP into future products rather than selling it as a standalone product line, similar to the acquisition strategy used for Mellanox [2][1] 2. **LPU Architecture**: The Groq LPU architecture is designed for ultra-low latency inference, particularly suitable for batch size=1 scenarios, complementing NVIDIA's GPU strengths in training and larger batch size inference [1][4] 3. **Timeline for Integration**: The integration of LPU into NVIDIA products is expected to take at least 18-24 months, likely aligning with the Finman generation of products [1][6] 4. **Chiplet Integration**: The LPU is expected to be integrated into the GPU architecture using chiplet technology, allowing for closer physical proximity to reduce latency [1][7] 5. **SRAM Utilization**: The LPU will utilize approximately 230MB of on-chip SRAM to minimize latency from external memory access, and its deterministic timing will enhance stable low-latency inference performance [5][4] 6. **Impact on HBM**: The integration of LPU is not expected to affect the usage of High Bandwidth Memory (HBM), as LPU's SRAM serves a different purpose within the memory hierarchy [8][1] 7. **Market Beneficiaries**: The industry chain likely to benefit from LPU integration will focus on triplet-related packaging rather than HBM or PCB-related areas [9][1] Other Important but Possibly Overlooked Content 1. **Interconnect Scalability**: The LPU has limitations in interconnect scalability, which NVIDIA plans to address by integrating LPU capabilities directly into the GPU architecture, thus avoiding interconnect issues between multiple LPUs [10][9] 2. **Software Integration**: There is a potential pathway for LPU's software system to be integrated into CUDA, allowing for unified memory management and scheduling [10][1] 3. **Upcoming GTC Focus**: The upcoming GTC is expected to highlight Rubin Ultra and Firemon architectures, with less emphasis on previously discussed topics [11][1] 4. **CPU Architecture Changes**: There is speculation that Rubin Ultra may introduce an X86 architecture option, reflecting a shift in market needs for inference processing [12][1] 5. **Intel's Strategy**: Intel may move towards a unified core strategy, focusing on power optimization, which could impact its competitive stance against AMD [13][1] 6. **Long-term AI Evolution**: The long-term view suggests that large language models (LLMs) are not the ultimate path to Artificial General Intelligence (AGI), indicating a need for new algorithms and approaches in AI development [16][1]