Videos

The gap between agent performance is driven by the harness (the loop around the model), not the model weights alone:...

A new research paper introduces NAPM Mem, a framework that transforms long-term user memory from passive retrieval...

Codex (ChatGPT's AI agent) can be mastered with 15 tips including browser use, voice mode, publishing sites,...

AMD's MI355X GPU matches or beats Nvidia's B200 on paper but loses by a third in benchmarks due to software issues...

A Tai Chi app for seniors generates $100K/month by targeting an underserved niche within the crowded fitness market....

Graph engineering is emerging as a targeted upgrade to RAG for complex, multi-hop questions, with recent benchmarks...

OpenAI's new model, Astra, reportedly solved 10 open problems in mathematics and theoretical computer science,...

The debate between loop engineering and graph engineering for AI agents is settled by the task's reliability...

Nate Herk shares 12 lessons from 5,000+ hours of building with AI, emphasizing that standing out requires...

Bijan Bowen tests Traycer, an open-source multi-agent orchestration tool, by having large models like Fable 5 and...

A new open-source model, Quen 3.8 Max from Alibaba, is competitive with top closed-source models like Fable and...

Inference engineering for large language models involves a stack of optimizations including KV cache-aware routing,...

AI's value shifts toward decision-making, arbitrage, private loops, and proof of work, not just task execution. The...

LM Studio's Bionic is a local agent that runs open-weight models on your machine, offering a free alternative to...

Simon Eskildsen, CEO of Turbopuffer, shares his journey from a self-taught programmer at Shopify to building a...

Alibaba released Qwen3.8 Max, a 2.4 trillion parameter mixture-of-experts model with 95 billion active parameters,...

The top 10 AI repos of July 2024 are all scaffolding around existing models, not new models themselves. Only two...

MCP Apps is an open protocol extension to the Model Context Protocol (MCP) that allows services to transmit their...

Tencent Cloud open-sourced an MIT-licensed memory plugin for AI agents that improves pass rates by 51% while cutting...

MCP tasks, designed for long-running async tools, are not yet widely adopted by agent clients because the V1...

David Ondrej open-sourced his personal repository of 42 agent skills for AI coding tools like Codex and Claude Code,...

Kimi K3 (2.78T parameters) beats DeepSeek V4 Flash (284B) on benchmarks by 4–6 points, but costs 21–54x more per...

OpenAI's GPT-5.6 Luna is now its cheapest model after an 80% price cut, costing $0.20 per million input tokens and...

Python dominates AI model training, fine-tuning, evaluation, and research, while TypeScript leads in AI product...

Waste is a 6,000-line C engine that runs the 2.78-trillion-parameter Kimi K3 mixture-of-experts model on a laptop...

Pricing AI automation systems should be based on the value delivered to the client, not on hourly rates. Nate Herk...

DeepSeek V4 Flash 0731 is a powerful local AI model that excels at reasoning and benchmarks but tends to overthink...

Teaching AI to hack effectively requires designing reinforcement learning environments with deterministic grading...

Long-horizon AI agent tasks are defined by both human-relative time and model-specific metrics like tokens and...

Current AI, built on RLHF, excels at assistance (human-in-the-loop tasks) but fails at true automation due to...

Jack Roberts demonstrates how to build AI agents that actually work using three levels: access, knowledge, and...

Claude Fable 5, combined with Higgsfield's MCP connector and Frame.io, can watch a video transcript, flag missing or...

Claude CoWork now runs on mobile with scheduled tasks executing in the cloud, but files remain local unless synced....

DeepSeek V4 Flash is a newly released official version of a small but powerful AI model with 284B parameters (13B...

QAC is a local AI SDK that lets you install and run multiple AI models (speech-to-text, embeddings, LLM,...

OpenAI dramatically cut prices on GPT-5.6 Luna by 80%, now at $0.20/M input and $1.20/M output, and used its own...

Richard Socher presents a vision for automating scientific discovery through a 'Eureka machine' that uses recursive...

Bottle Cap AI's ThinkingCap fine-tune of Qwen 3.6 27B reduces reasoning tokens by ~46% while preserving benchmark...

Abacus AI's supercomputer, using GPT-5.6 Sol and Fable 5 in max mode, autonomously built and deployed a full-stack...

Production bugs in finance agents are often caused by slow developer loops, not model or hardware limitations. The...

Graph engineering applies graph theory to orchestrate multiple AI agents in dynamic workflows, enabling complex...

AI finance products are built to impress for five minutes but fail under real scrutiny because they prioritize...

The Open Knowledge Format (OKF) provides a universal standard for building AI-readable knowledge bases from YouTube...

Integrating AI agents into event-sourced systems can handle uncertain cases like real-time fraud detection by using...

Deploying AI agents in group settings (e.g., family, work chats) introduces unique challenges around security,...

Nubank built a security review system to vet AI skills before they reach developers, treating them as supply chain...

Kepler built verifiable AI for financial services by augmenting large language models with a deterministic substrate...

Synthetic personas, powered by LLMs, are emerging as a tool for market research to simulate human respondents, but...

Off-the-shelf LLMs give fluent but unreliable financial advice because they have read about money but lack...

Nubank ships AI agents 20× faster by using simulated eval data instead of waiting on production traces. Generating...