Have you heard these exciting AI news? - August 22, 2025 AI Updates Weekly

Have you heard these exciting AI news? - August 22, 2025 AI Updates Weekly Presented by Lev Selector Slides - https://github.com/lselector/seminar/tree/master/2025 (click on pptx file, then on "raw" or download button on the right) # Please help Please write in comments if you have experience using sub-agents and hooks with Claude Code # Need AI Consulting ? - Enterprise AI Solutions - https://EAIS.ai - Linkedin - https://www.linkedin.com/in/levselector - GitHub - https://github.com/lselector # Our books: "Artificial Intelligence In Business: Strategies That Work" https://www.amazon.com/dp/B0FKZ3364G/ "AI Product Manager: Building Tomorrow" https://www.amazon.com/dp/B0FMS9NFK9/ # Contents of today's video: 00:00 Trend: More Memory & More Personalization 00:38 Crowd-sourced "LM Arena" Leaderboard 01:04 Google A2A vs MCP 02:03 World Humanoid Robot games in Beijing 02:34 Google LangExtract 03:17 NVIDIA Audio Canary-1b & Parakeet-tdt-0.6b 03:39 Nano Banana Text-to-Image (Google ?) 04:20 Archon - MCP - shared memory, project context 05:17 Six Micro-Tools Generating Passive Income 05:40 Google Gemini URL Context tool in Gemini API 06:54 A solo programmer built web search engine 07:18 GPT-5 - 95.84% accuracy on MedQA 07:39 Microsoft ads GPT-5 into all products 07:57 Qwen-Image-Edit 20B open-source model 08:49 90%+ of game developers use AI 09:00 95% of genAI pilots at companies are failing 09:27 DuckDB vs SQLite vs Pandas vs Polars 10:14 DeepSeek v3.1 - better at Coding 10:37 Anthropic is raising $10B 11:16 Pixel 10 rollout Aug 20 - AI Features 12:02 Meta DINOv3 - Self-supervised learning for vision 12:21 Gemini API - Imagen 4 Fast and Imagen 4 family 12:27 FastAPI-MCP 12:54 Alibaba wan.video - Wan 2.2 - open-sourced 13:07 ByteDance Seed-OSS 36B LLM 13:33 DynamicsLab Mirage 2 14:21 Thyme: Think Beyond Images 15:11 Claude Code - output Styles, Templates 18:59 Agents.md standard 19:36 Using Windmill.dev for AI Pipelines 20:33 GPT-5 Prompt Optimization Guide 21:28 The State of AI 2025 - BVP Report 22:48 Microsoft POML 23:31 HealthChain 24:27 Jobs, Layoffs # Takeaways 1. AI is becoming more personalized and context-aware.** The central trend is the push for models to have more memory and personalization, allowing them to understand user prompts and context more deeply, much like a human with memory. 2. AI capabilities are rapidly advancing across specialized domains. New models like GPT-5 are achieving near-perfect accuracy in medicine, while others like Qwen-Image-Edit and Nano Banana are setting new standards in image generation and editing. Coding assistants are also evolving from simple tools to complex multi-agent systems. 3. The business application of AI is maturing, with clear successes and failures. While a high percentage of generative AI pilots are failing (especially in sales and marketing), AI is finding significant success in automating back-office tasks, improving efficiency, and is being heavily adopted in industries like game development. # Quotes "The trend for this week is more memory and more personalization... a human without memory cannot act reasonably... Same actually with models. They need memory and they need personalization so that they know what you mean in your prompts." "95% of generative AI business pilots are failing and they are failing not because of technology but because they usually applied to sales and marketing where they don't really produce very good results but they actually produce best results from automating back office tasks." "Working with a code assistant is not just writing a small procedure... you ask the system to create a plan which consists of multiple steps and then for different steps you create configurations for different agents and then starting with planning you start executing those steps." # Summary Key trend - adding more memory and personalization to AI models to make them more effective. Updates in the LLM space - latest leaderboard rankings for GPT-5, Gemini, and Claude, dive into powerful new open-source models for image editing, audio transcription, and coding from companies like Google, NVIDIA, and Alibaba. New protocols and standards for AI agents, practical tools for developers - Claude Code and Windmill.dev. Business side of AI - why many corporate pilots are failing while adoption in areas like game development soars. # Keywords AI Agents, Personalization, GPT-5, Google, Microsoft, Anthropic, Claude Code, Open Source, Multi-agent Systems, Generative AI, LLM, Nano Banana, Prompt Optimization, Gemini API, NVIDIA, Qwen-Image-Edit, BVP Report