Have you heard these exciting AI news? - September 12, 2025 AI Updates Weekly

Have you heard these exciting AI news? - September 12, 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 write in comments how would you build a RAG for a client today?* Will you buy a product? Or use a framework? ------------------------------------------------------- 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 OpenAI: "avoid hallucinations by abstaining when unsure" 00:26 OpenAI on Hallucinations 02:35 Crowd-sourced "LM Arena" Leaderboard 03:10 Geoffrey Hinton warns about LLMs' Manipulations 03:28 Thinking Machines Lab Blog 04:08 K2-Think 32B & Baidu ERNIE-4.5-21B-A3B 06:19 Claude works with MS Office and PDFs 07:38 Microsoft Shifts Office 365 AI to Anthropic Claude 07:44 Peter Diamandis about AI 09:51 ByteDance Seedream 4.0 Image Model 10:19 Databricks secured $1B at $100B valuation 11:15 The Rundown AI - life-changing prompts 12:31 Google Veo 3 and Veo 3 Fast 12:43 Mistral Valuation $14B 13:24 Oracle's stock surged by 40% Wednesday 14:13 Cognition Raises $400M at $10.2B Valuation 14:44 MCP Registry 15:00 The Training Imperative 15:28 AI Gross Margins in 2025 15:48 Multi-word prediction 16:16 Moonshot AI Kimi-K2-Instruct-0905 16:37 ChatGPT supports integration with MCP Servers 16:59 Replit Raises $250M at $3B Valuation 17:55 ByteDance mini-03 18:22 Claude AI Outperforms Human Virologists 18:41 Nvidia Rubin CPX GPU handles 1 Mln context 19:15 CODE BUFF AI Agentic Coding System 20:18 InternVL3.5 open-source multimodal LLM 21:55 Alibaba 24:49 Meta REFRAG 26:44 Agentic AI Patterns Book 27:14 Claude Code - YouTube Research Agent 28:38 Customer Service AI Agent 29:08 Streamlit-Based RAG Debugging 29:13 HyperGraphRAG 30:11 GenAI Assessment Framework (GAF) 32:10 Stop Writing Tests 32:51 Tauri + Rust + Svelte for Native Apps 33:57 Synechron AI Consulting - $1B Revenue 35:05 Andrew Ng - Demand for AI Developers 37:09 Wes Roth: "AI just took all our jobs" 38:48 Jobs, Layoffs ### The 3 Main Takeaways 1. *AI's Shift from Bluffing to Abstaining:* The traditional training methods for AI models have encouraged them to "bluff" or provide a confident but incorrect answer rather than admitting they don't know. To combat this, OpenAI is advocating for new benchmarks and training methods that penalize confident errors more heavily and reward abstaining when unsure, which reduces hallucinations. 2. *The Rise of Agentic AI and Multi-Agent Systems:* AI development is moving beyond simple code assistants to more autonomous, multi-agent systems. Projects like *CodeBuff AI* and *Replit's Agent 3* demonstrate a future where AI agents collaborate, plan, and execute complex, multi-step tasks autonomously. 3. *The New Imperatives for Businesses and Professionals:* Businesses are finding more success by *buying existing AI solutions* rather than building them from scratch. Meanwhile, the job market is seeing a talent shortage for AI specialists, even as entry-level jobs in other fields decline. The future demands that developers and professionals *reskill* to work with AI building blocks like RAG, agents, and prompting to stay relevant and highly productive. --- ### Top 3 Quotes * "Avoid hallucinations by abstaining when unsure." - *OpenAI* (This quote, also the video's epigraph, summarizes a key problem in current AI and its proposed solution.) * "We are moving from someone with technical skills can create a fake video to anyone with 20 bucks can generate thousands of fake videos in an afternoon with a very high quality." - *Peter Diamandis* (This quote highlights the rapid democratization of powerful, and potentially dangerous, AI capabilities.) * "The best developers I know aren't recent graduates. So he says that computer science knowledge is still essential, but with AI, people can become much more productive." --- ### Short Summary Latest trends and updates in artificial intelligence. Focus on OpenAI's new approach to reduce AI hallucinations by encouraging models to abstain from answering when they are unsure. Major funding rounds for companies like *Databricks* and *Mistral*, and Oracle's significant growth driven by its AI data centers. The shift toward multi-agent AI systems and the growing demand for developers who can work with these tools. Changing job market, noting the rising need for AI specialists and the importance of re-skilling to adapt to this new era of technology.