🚀 Apparently TSA missed the “3-oz limit on AI models” in my carry-on. #dadjoke #staywithme Thirty days after joining AWS, OpenAI’s brand-new open-weight models, gpt-oss-120b and gpt-oss-20b, just landed in Amazon Bedrock and SageMaker. Total coincidence…

Finished the San Francisco Marathon, and set a new PR:4 h 20 m! 🎉 Two years ago I stood on these same streets with a race bib and a busted leg, watching everyone else start while I sat on the sidelines. It hurt 😞 literally and figuratively. I’d trained, I’d planned, and life still said “not today.”

💡 Bigger isn’t always better: why I’m reaching for smaller models in my everyday work When I first got access to O3, its raw power blew me away, I honestly thought I’d never settle for anything less.

On Saturday, August 16 I’m trading weekend plans for something way more fun: the DeepSeek Demystified Summit, a one-day, all-access dive into the open-source LLM that’s making waves across cost, performance, and reasoning. Why I’m excited:

Text-diffusion LLMs just graduated from lab demo to production reality. Let me show you why that matters… 👇 Inception Labs’ brand-new Mercury chat model streams ≈708 tokens /sec on ONE H100, \~7× faster than GPT-4-class “speed” variants on identical hardware. ⚡️

🤯 What if your Transformer paid attention to triangles instead of lines? Those extra corners buy you token efficiency and a steeper scaling curve.

Just wrapped up my first week with the Bevel Health app, and I’m honestly blown away. 📱✨

🚀 Counting down to European AI and Cloud Summit next week! I’m excited to head to Düsseldorf for the European AI & Cloud Summit (26-28 May) where I’ll be taking the stage to share some fresh insights on Multi-Agent AI Workflows.

🤯 What if AI agents could sense when they’re out of their depth—pausing to reflect or fetch knowledge before acting? Meet “KnowSelf,” a new paradigm for LLM agents. Thread below 👇

🚀 Large language models excel at facts, but can they craft plans for brand-new tasks? Introducing Analogy-Augmented Generation (AAG)! #AIResearch #LLM

⚡️ OpenAI’s o3 & o4-mini are here, moving us from chatbots to autonomous analysts. These models browse the web, run Python, process images, and plan actions. #AI #OpenAI

🚀 OpenAI just launched GPT-4.1, GPT-4.1 mini, and GPT-4.1 nano – a new family of models with stellar performance upgrades. Available now via API only, not in ChatGPT yet! #ArtificialIntelligence #GPT4.1

🧬 Google DeepMind's TxGemma doesn't just predict drug molecules—it EXPLAINS why they might work! This could unlock a major bottleneck in AI-powered drug discovery. #AIinHealthcare #DrugDiscovery

After nearly 3 years away from racing due to injuries, I'm finally back on the course — and it feels amazing! 😊 Ran my first 10K since the break and finished 12th out of 300 in my age group. Not a bad comeback! 🏃‍♂️💨

Why I Refused to Play the AI Detection Game. I walked away from a publisher over their obsession with "AI detection" thresholds. They wanted me to tweak my original work to dodge some arbitrary % limit. I’d rather focus on quality than dance for a tool.

What if neural networks were a Wonderland, and we're all just curious Alices trying to make sense of it? Just finished Scardapane's brilliant 'Alice's Adventures in a Differentiable Wonderland' and I'm still processing how elegantly it demystifies the rabbit hole of neural networks.

Netflix is revolutionizing recommendation systems by applying Large Language Model (LLM) principles at scale. Their foundation model processes hundreds of billions of user interactions—comparable to the token volume in ChatGPT. #AIforRecommendation #FoundationModels

LLaMA 4’s 10M token context window sounds wild—bye-bye vector databases and RAG? Not quite. The hype is big, but the story’s more complicated. Let’s break it down. 🧵 #LLMs #AI

The last time I took the Scrum Master I exam was back in 2012 😊 Yesterday, I figured I’d give it a shot—and I passed Scrum Master II ! Turns out, I really didn’t need to wait this long. The questions were solid. I’d say anyone with 3–5 years of Scrum experience should be able to pass pretty easily.

Not every AI problem needs a multi-agent system. Most don’t. Chasing complexity too early wastes time & burns trust. #CTO #GenAI #MultiAgentSystems

🤯 72% of professors fear AI lets students skip thinking. But Claude for Education’s new Learning Mode flips the script—designed to spark learning, not shortcut it. #AIinEducation #EdTech

🚀 BREAKING: A paralyzed woman speaks again after 18 years—using only her thoughts! 🧠✨ Her own voice, powered by a brain implant & AI, communicates in real-time. Not sci-fi, but science. 👇

🚨 Big news in mental health: An AI therapist achieved human-level therapeutic alliance in half the time, reducing depression by 51%, anxiety by 31%, and eating disorder concerns by 19% in a groundbreaking clinical trial. 🧵

🧠 Can an 8B open-source model outperform GPT-4 in collaborative tasks? Turns out—it can. But only with the right training recipe. 👇

🚀 MCP: The Protocol Poised to Revolutionize AI Tool Integration Imagine a world where AI agents autonomously select and sequence tools without manual coding. That’s the promise of the Model Context Protocol (MCP)—a new foundation for AI workflows.

Was deep into prompt engineering this morning when I stumbled upon OpenAI's latest: the 4o Image Generation model. It's incredible! Here’s why: #AIImageGeneration #OpenAI #GPT4o #PromptEngineering #AITechnology

🚀 Anthropic just dropped a major upgrade for Claude: the "think" tool! This lets Claude pause, reflect, and tackle complex, multi-step tasks with ease. Developers integrating LLMs into apps, this is for you—here’s why it’s a big deal. #AI #LLM #DeveloperTips

🔍 In high-stakes scenarios, wrong AI answers can be costly. New research from Johns Hopkins shows "test-time scaling" boosts accuracy and helps models know when to stay silent.

Hey everyone! 🚀 OpenAI just dropped some insane new tools for voice agents. I’m breaking it down for you—let’s dive into the highlights! 🧵

Another day, another agentic framework! NVIDIA's AgentIQ treats agents, tools, and workflows as simple function calls for true composability. Build once, reuse everywhere. 🔄 #AIEngineering #LLMOps

Welcome Eclipse LMOS - an open-source, cloud-native platform laying the groundwork for an "Internet of Agents." 🌐 Think of it as connecting AI agents across orgs, just like the Internet linked computers and IoT tied devices. #AIAgents #OpenSource

Baidu just dropped ERNIE 4.5 & ERNIE X1, pushing AI to new heights! 🚀 ERNIE 4.5 rivals GPT-4.5 in benchmarks but at a fraction of the cost. ERNIE X1 excels in logical reasoning & problem-solving. Big news for AI devs & researchers! #AI #MachineLearning #NLP

What if your AI assistant could not just remember conversations, but actively organize and evolve its memories like a human? New research from Rutgers & Ant Group introduces "A-MEM" - and the results are remarkable. 🧠 🧵 Thread on agentic memory for LLMs

🔥 NEW: Google DeepMind just released Gemma 3, and it's a game-changer. Their 4B parameter model now matches last year's 27B model performance while adding vision capabilities and 128K token context windows. All on consumer hardware! 🤯 Thread 🧵👇

🔍 "Let's hack the tests instead." When OpenAI researchers examined their frontier AI models, they found them openly discussing how to cheat rather than solve problems correctly. Yet when they trained models not to say such things, the models continued cheating—they just stopped talking about it.

🚀 OpenAI just launched their Agents Platform! This is the big drop we’ve been waiting for to make 2025 "The Year of the Agent." #AIEngineering #OpenAI

🚀 Introducing OpenManus: The open-source alternative to Manus AI that lets you build autonomous agents WITHOUT invite codes! MetaGPT researchers launched a working prototype in just 3 hours. Here's why this matters...

🚀 Introducing OpenManus: The open-source alternative to Manus AI that lets you build autonomous agents WITHOUT invite codes! MetaGPT researchers launched a working prototype in just 3 hours. Here's why this matters...

🚀 Introducing OpenManus: The open-source alternative to Manus AI that lets you build autonomous agents WITHOUT invite codes! MetaGPT researchers launched a working prototype in just 3 hours. Here's why this matters...

🚀 Introducing OpenManus: The open-source alternative to Manus AI that lets you build autonomous agents WITHOUT invite codes! MetaGPT researchers launched a working prototype in just 3 hours. Here's why this matters...

Are you still thinking at the token level? Large Concept Models (LCMs) are changing how AI understands language, moving beyond word-by-word processing to unlock more human-like reasoning capabilities.

So many people have been asking me for math resources 😊 I can't even count. I found another amazing set for those who want to start everything from the beginning. The videos come from a high school teacher who taught math for 29 years. Maybe you can watch them with the family 😊

Ever wondered if LLMs can think too much? New research shows that maintaining excessive reasoning history actually HARMS performance. Less can be more when it comes to AI reasoning! 🧠

🔍 What if solving GitHub issues could teach language models to think better? Meta's new research shows a 41% solve rate on real-world software bugs while unexpectedly boosting performance across unrelated domains. A thread on SWE-RL and its surprising findings 🧵👇

Latest and greatest, as soon as possible 😊 I love the execution speed @github.com .

🧵 Ever wondered how to orchestrate multiple AI "agents" to tackle complex tasks? I recently delivered a talk on building "Agentic LLM Workflows with Autogen ," going from simple setups to advanced multi-agent collaboration. Here are key takeaways (and the video link at the end):

Did you know LLMs can solve complex problems using just 7.6% of the tokens while maintaining the same accuracy? The assumption that verbose reasoning is necessary for AI problem-solving just got challenged. 🧵

🔥 Just released: Complete AI Engineer Summit recordings from OpenAI, Anthropic, Google & 50+ top AI companies! Free insider knowledge that would've cost you a conference ticket. An absolute goldmine for anyone in AI. 🧵👇

"Is your AI knowledge keeping pace with today's rapid innovations?" Foundation models are reshaping our digital world—and MIT is offering front-row seats to understand how. 📚 MIT Introductory Course on Foundation Models & Generative AI

🧠❓ Did you know AI is learning to cheat when it's losing? Recent studies show advanced models like OpenAI's o1-preview attempting to hack their opponents in chess matches! #AIEthics #MachineLearning