LearnByDoing With Steven

•

Канал • Подкастов: 4

Образование

Образование

Еженедельно
Образование

Образование

Еженедельно
Образование

Образование

Еженедельно
Образование

Образование

Еженедельно

-3 дн.

🚀 Stanford CS336 Lecture 13: The Evolution of Language Model Data. -Notebooklm Summary

Stanford CS336 Lecture 13 focuses on the critical evolution of language model training data. While model architectures are widely disclosed, dataset details remain highly proprietary due to commercial competition and copyright considerations. The lifecycle of language model training spans pre-training, mid-training, and post-training. The mid-training phase curates high-quality datasets to enhance specific capabilities like coding, mathematics, and long-context reasoning. Ultimately, data curation shifts from high-volume, low-quality web crawls to low-volume, high-quality specialized datasets. Data filtering methodologies have evolved from basic rule-based heuristics and language identification to advanced model-based approaches. Modern curation pipelines leverage large language models to assess the educational value of documents, rewrite low-quality texts, and synthesize high-quality QA pairs to scale data efficiently. Legal and compliance challenges, including copyright and fair use, remain central to data acquisition. As models risk memorizing training text, developers navigate the balance between direct commercial licensing and fair use arguments. Key Takeaways: Mid-training acts as a crucial bridge, refining models for targeted reasoning tasks.Advanced LLM-driven filtering and synthesis scale high-quality data while avoiding rule-based bias.Copyright compliance and memorization concerns limit public dataset disclosures.All my links: https://linktr.ee/learnbydoingwithsteven #learnbydoingwithsteven #AI #DeepLearning #Research #TechSummary #MachineLearning #LLM #DataScience #DataCuration #Copyright #ArtificialIntelligence

-3 дн.

•

7 мин.
16 янв.

The $20 Billion Inference Pivot: Nvidia’s Strategic Groq Acquisition

This document analyzes Nvidia’s $20 billion acquisition of Groq, a strategic move designed to dominate the rapidly growing AI inference market. By absorbing Groq’s specialized Language Processing Unit (LPU) architecture, Nvidia aims to overcome traditional hardware bottlenecks and deliver the ultra-low latency required for real-time autonomous agents. The report highlights how this deal serves as both a defensive moat against rival chipmakers and a proactive pivot toward a future where applying AI models becomes more profitable than training them. Furthermore, the analysis explores the regulatory complexities of the "asset-plus-talent" deal structure and its potential to reshape the global AI hardware landscape. Ultimately, the text presents the acquisition as a high-stakes gamble to ensure Nvidia remains the foundational operating system of AI infrastructure.

16 янв.

•

15 мин.
16 мар.

Beyond the Super Agent: Designing Collaborative Agentic Systems

Modern AI development is shifting away from the "super agent" myth toward collaborative systems where multiple specialized agents work together on specific tasks. To ensure safety and efficiency, developers must avoid over-privilege and "super agency" by enforcing high cohesion, which limits each agent's access to the absolute minimum required for its role. This architectural approach categorizes agents within a two-by-two matrix that measures risk versus capability, ranging from low-stakes information retrieval to high-consequence financial transactions. High-risk, high-capability agents require ephemeral existence, dynamic access, and human-in-the-loop oversight to prevent unintended damage. By treating agents as a spectrum of capabilities rather than a one-size-fits-all solution, organizations can build smarter, more secure systems that remain under human control.

16 мар.

•

48 мин.
18 июн.

Stanford CS336 Language Modeling from Scratch Lecture 12 highlights - Evaluation Overview

Stanford CS336 Language Modeling from Scratch Lecture 12 Evaluation Overview Evaluating language models may seem as simple as measuring a specific model's performance, but it is actually fraught with challenges. The industry currently evaluates models through various metrics, such as benchmark scores like MMLU, cost-effectiveness indicators combining model accuracy and per-token cost, OpenRouter platform data based on user traffic routing, and Chatbot Arena which relies on human pairwise preference comparisons. However, an evaluation crisis currently exists, as some benchmarks may have reached saturation or been gamed, making it difficult to determine the most accurate evaluation method amidst a plethora of models and benchmark data. Key Takeaways: The fundamental purpose of evaluation depends on specific needs, and there is no single true evaluat...All my links: https://linktr.ee/learnbydoingwithsteven #learnbydoingwithsteven #AI #DeepLearning #Research #TechSummary #MachineLearning #LLM #ScalingLaws #NeuralNetworks #Innovation

18 июн.

•

5 мин.
17 мар.

OWASP Top 10 Vulnerabilities for Large Language Models

This source outlines the OWASP Top 10 vulnerabilities specifically targeting Large Language Models (LLMs) and AI systems. It highlights critical risks such as prompt injection, where attackers manipulate model instructions, and sensitive information disclosure, which can lead to data breaches. The text also warns against supply chain vulnerabilities and model poisoning, emphasizing that the data and third-party models used to build AI must be rigorously vetted. To combat these threats, the author recommends implementing AI firewalls, enforcing strict access controls, and performing regular penetration testing. Ultimately, the overview serves as a guide for developers to prevent excessive agency and system malfunctions in production environments.

17 мар.

•

45 мин.
10 янв.

LLM-D: Optimizing Distributed AI Inference with Intelligent Routing

The provided text introduces LLM-D, an open-source project designed to optimize AI inference by treating requests like planes managed by an air traffic controller. By utilizing a distributed architecture on Kubernetes, this system significantly reduces latency and operational costs for complex tasks like RAG and agentic workflows. It functions through an inference gateway that intelligently routes prompts based on current system load and the likelihood of cached data. A key innovation is the disaggregation of the prefill and decode phases, allowing them to scale independently while sharing resources. Ultimately, this approach achieves massive improvements in response times and throughput, making it ideal for high-demand, mission-critical AI environments.

10 янв.

•

12 мин.
20 февр.

The 2025-2026 AI Technology Stack Panorama Report

The provided text outlines the comprehensive evolution of the global AI technology stack between 2025 and 2026, transitioning from experimental tools to integrated autonomous systems. It details a shift toward specialized hardware and advanced cooling solutions like liquid cooling to manage the immense power demands of modern computing. The report emphasizes the critical role of refined data engineering and architectural shifts, such as the rise of sparse Mixture of Experts (MoE) models, to improve efficiency and reasoning. Beyond infrastructure, it explores the optimization of inference engines and the rise of intelligent agents capable of multi-step planning and tool usage. Furthermore, the sources contrast divergent global regulatory landscapes, highlighting the strict risk-based approach of the European Union against the innovation-focused policies in the United States. Ultimately, the documentation predicts a future where AI becomes deeply embedded in operating systems and physical robotics, eventually leading to self-evolving research and development.

20 февр.

•

17 мин.
9 мар.

Steven Data Talk EP16[Notebooklm Ver] – Conversation with Tianze Tang, PhD Candidate at the Medical University of Vienna – Study Methods, Research Practice, AI and Research, Large Language Models, AI

Steven Data Talk EP16 – Conversation with Tianze Tang, PhD Candidate at the Medical University of Vienna – Study Methods, Research Practice, AI and Research, Large Language Models, AI Development Trends, and Academic Planning & Higher-Education Advice in the AI Era Tianze Tang joined the Medical Data Science Department at the Medical University of Vienna in September last year as a PhD candidate. His primary research focuses on medical AI for retinal imaging. He received his Bachelor’s degree in Biotechnology from Xi’an Jiaotong University and his Master’s degree in Biostatistics from New York University. He has explored and developed knowledge across nearly all areas of AI. Personal website: https://ttzaiweb.com/ GitHub: https://github.com/TianzeTang0504 My link: https://linktr.ee/learnbydoingwithsteven Release date: March 9, 2026 (please refer to the social media matrix)

9 мар.

•

7 мин.
27 апр.

How to build "Professional Grade" Agent Skills? 8 Essential Tips. 🛠️

How to build "Professional Grade" Agent Skills? 8 Essential Tips. 🛠️ 1️⃣ Precision Triggers: Description determines success. Too broad leads to misfires; too vague leads to failure. 2️⃣ Constraints vs. Paths: Define the goal and give the agent space to iterate, avoid rigid SOPs. 3️⃣ Layered Loading: Keep the core instructions lean. Load auxiliary files on demand to save context. 4️⃣ Negative Cases: Defining what NOT to do is as critical as defining what to do. 5️⃣ Retirement Plan: If the base model absorbs the capability, delete the skill. A great skill’s ultimate goal is to become obsolete. All my links: https://linktr.ee/learnbydoingwithsteven #learnbydoingwithsteven #AgentSkills #AI #SoftwareEngineer #PromptDesign #LLM #Automation #TechTips #FutureOfTech #DeveloperGuide

27 апр.

•

7 мин.
17 февр.

Intelligent AI Delegation: A Framework for Agentic Coordination

The provided text explores intelligent delegation, defined as a structured framework for assigning complex tasks between humans and AI agents while maintaining accountability and trust. To ensure safety, the process involves task decomposition, where objectives are broken down into verifiable sub-tasks based on their difficulty and importance. The authors propose a dynamic coordination cycle that allows for real-time adjustments, such as switching assistants if performance drops or environmental conditions change. Success is supported by reputation systems and smart contracts that formalize agreements and automate penalties to ensure reliability in decentralized markets. Additionally, the research highlights the need for transitive monitoring and cryptographic verification to prevent security threats like data poisoning or unauthorized access. Ultimately, the framework aims to scale autonomous systems effectively by balancing operational efficiency with human oversight and ethical labor standards.

17 февр.

•

17 мин.

Learn By Doing With Steven 数能生智