Everything tagged machine-learning, newest first. Tags come from the classifier reading each item's summary; 394 tags used 25 times or more have their own page.
Netflix describes how multimodal embeddings from CLIP and MediaFM improved asset personalization by solving cold-start problems, consolidating artwork models across canvases, enabling query-aware search artwork, and boosting video preview ranking, all supported by a shared Embedding Store and validated through offline IPS and online A/B tests.
Mastercard and McLaren leaders discussed how AI and data science have evolved from experimental efforts into core, trusted capabilities, with McLaren’s engineering culture relying on rapid iteration, data-driven car development, and acceptance that not every AI project will work immediately.
Perceptron presents an AI model designed to help machines navigate their environment while delivering detailed visual intelligence for better perception and decision-making.
Cisco and Basler are collaborating on a validated machine vision design that connects cameras, industrial networking, and AI-driven decision-making to deliver reliable, scalable vision data transmission across factory and other industrial environments.
Medium explores the use of Large Language Models (LLMs) for text classification, particularly for updating their NSFW model, finding it effective but not suitable for all models due to cost and adaptability concerns.
This tutorial guides you through building a custom reinforcement learning library in C, implementing an autograd engine, and coding a snake game environment to train an AI agent using the policy gradient reinforce algorithm.
Flock Safety, an $8.4 billion surveillance startup, uses automated license plate readers to create a nationwide vehicle tracking database, exploiting legal loopholes and raising privacy concerns.
OpenAI is introducing a faster version of its latest, most advanced model to attract enterprise clients.
AI systems, like GPS, can confidently provide incorrect information, known as hallucinations, which can be mitigated by grounding agents with verification tools, though they still pose risks when taking autonomous actions.
Analyzing Anthropic's results is challenging, but an attempt is made to discern valuable insights from the data.
Google's Gemini team is actively developing the Gemini 3.5 Pro and Gemini 4 models, with recent checkpoints showing promising results, while competitors like Enthropic and OpenAI have raised industry standards with their advanced models.
Michael Foree, Stack's Director of Data Science, educates Phoebe on AI context, context engineering, and strategies to improve her skills as a context engineer.
Quantum machine learning combines classical machine learning with quantum computing principles, utilizing qubits and entanglement to enhance computational capabilities and achieve parallelism, enabling more complex data processing and potentially offering advantages over classical systems.
Integrating AI into systems requires strategic decisions, balancing trade-offs between human judgment, rule-based logic, machine learning, and generative AI, each suited for specific tasks with their own advantages and limitations.
Mera Miati, former CTO of OpenAI, founded Thinking Machines, which released Inkling, a unique open weights model that prioritizes efficiency and epistemic humility over raw intelligence, despite being overshadowed by larger models like Kimmy K3.
The AI industry is rapidly advancing with Chinese AI labs, including Alibaba, preparing new models like Quen 3.8 and Quen 4.0, while other companies like Deepseek and GLM are also testing and expanding their AI capabilities, amid a growing ecosystem of tools like Zapier SDK to enhance AI integration.
Netflix has developed an in-house LLM serving platform using vLLM and Triton to integrate machine learning models directly into their production environment, focusing on engine selection, model packaging, API design, and deployment strategies to optimize performance and flexibility.
The 2002 paper "Language Trees and Zipping" demonstrates how file compression, specifically using gzip, can cluster languages and reveal their lineage by measuring co-compression distances, linking compression theory with machine learning concepts like cross-entropy, which is fundamental in training modern language models.
OpenAI has launched the GBT 5.6 family, including Soul, Terra, and Luna, with Soul setting new standards in intelligence and efficiency across various domains, outperforming previous models while being faster and more cost-effective.
GPT 5.6, OpenAI's latest model, excels in complex tasks with three variants, offering advanced capabilities like real-time interactive web apps and physics simulations, best utilized with the Codeex desktop app for multi-agent tasks.
The shift from signature-based detection to machine learning and autonomous agents transforms cybersecurity by evaluating if actions make contextual sense rather than matching known patterns.
Google Deep Mind is facing internal challenges but aims to regain momentum with the upcoming Gemini 3.5 Pro model, featuring a new base model, expanded context window, and advanced reasoning capabilities, set for tentative release on July 17th.
Recent AI advancements include a full-body waifu, a Chinese model trained without Nvidia GPUs, global availability of Claude Fable 5, efficient Google models, real-time video editing, thought-to-text conversion, and a new sheet music reading AI called Musev VIT.
AI code review leverages machine learning and large language models to automate and enhance the code review process, improving consistency, reducing technical debt, and providing immediate feedback for developers, thereby accelerating workflows and reducing human error.
OpenAI introduces three new models, GBT 5.6 Soul, Terra, and Luna, with Soul Ultra leveraging sub-agents for complex tasks, surpassing competitors in benchmarks like terminal and exploit bench, highlighting cost-effectiveness and improved reasoning capabilities.
GLM 5.2 is the latest state-of-the-art model accessible via chat.zai.com, offering advanced features like search-enabled chat and agent areas for varying task complexities, including research, visualization, and basic coding.
The concept of a "data science periodic table" organizes data science elements into rows and groups, illustrating their roles in the analytics lifecycle from raw data extraction to evaluation, helping decode and build data science systems.
Netflix is exploring AI-driven video editing tools, Vera and VOID, to enhance creative control by preserving original footage integrity while enabling precise edits, addressing challenges like unintended edits and unnatural physics in current models.
From Alan Turing's 1936 exploration of algorithms to Claude Shannon's 1948 bit invention, foundational computer science papers have shaped today's technology landscape, influencing everything from AI to neural networks.
Cloudflare is enhancing its AI capabilities by integrating Ensemble AI team members to improve machine learning infrastructure and efficiency.
The emergence of AI has introduced numerous new terms and slang, necessitating a glossary to understand key words and phrases.
Graph Neural Networks (GNNs) effectively model complex, interconnected data by using nodes and edges to represent entities and their relationships, utilizing message passing to aggregate information and generate embeddings for both heterogeneous and homogeneous graphs.
Google's Gemini 3.5 Flash, launched at the Google IO conference, is a fast, efficient, and strong coding model with impressive performance but high token costs, raising questions about its efficiency as a flash tier model.
Google's upcoming IO conference will showcase the impressive Gemini 3.5 Flash and Pro variants, featuring remarkable improvements in AI-generated content, including a Minecraft clone with enhanced aesthetic and dynamic capabilities.
The content explains a method to reverse-engineer image prompts using ChatGPT Images 2.0, enabling users to recreate and learn from existing images by analyzing them and generating detailed prompts.
Thinking Machines aims to revolutionize AI interaction by developing a model that simultaneously processes input and generates responses, creating a more conversational experience akin to a phone call.
Machine learning and autonomous agents are transforming detection methods by moving from pattern recognition to contextual understanding.
ChatGPT's hallucinations have been reduced by 52%, especially in specific domains like medicine, law, and finance, improving accuracy in providing confidently presented facts.
Predictive AI forecasts specific outcomes using structured data, while generative AI creates new content from unstructured data, serving distinct purposes with different outputs and data requirements.
As we increasingly interact with computers through voice communication, work setups will evolve to accommodate new technologies, enhance productivity, and improve user experience.
Netflix has transformed its machine learning infrastructure to support diverse business domains, overcoming fragmentation by implementing a Metadata Service and Model Lifecycle Graph, enabling cross-domain collaboration, discovery, and exploration of ML assets.
GPUs, initially designed for gaming, have become essential for generative AI due to their superior ability to handle large-scale computations, unlike traditional CPUs, which are less efficient for AI tasks due to their general-purpose design.
OpenAI's GPT 5.5 is a revolutionary model that excels in efficient task completion, coding, research, and data analysis, using significantly fewer tokens and offering state-of-the-art performance across various benchmarks.
The process of inputting prompts and generating outputs involves hidden decision-making steps that are not visible to users.
AI agents require a team of specialized roles, similar to human teams, to collaboratively tackle complex tasks by breaking them down into manageable parts, planning, executing, and integrating external knowledge.
OpenAI's new model, GPT 5.5 Pro or GPT6 (also known as GPT Spud), promises significant advancements in understanding and solving complex problems, having been developed over two years with a new base model, and is already being previewed by users who report it as highly capable and user-friendly.
The job of a prompt engineer has evolved into a complex role requiring a broad skill set beyond crafting prompts, encompassing system design, API management, machine learning, and more to build AI agents that function effectively in real-world applications.
Dave demonstrates training a neural network on a vintage PDP144 computer, highlighting the fundamental principles of neural networks and the role of transformers in understanding sequence data through a simple task of reversing an eight-digit sequence.
AI development is plagued by technical debt due to rushed deployments and lack of planning, leading to future costs and system fragility, with strategic and reckless debt differing in awareness and preparedness.
Arcee, a small U.S. startup with 26 employees, has developed a large, high-performing open-source LLM that is becoming popular among OpenClaw users.
