ThursdAI - The top AI news from the past week

This week features Todd Segal, a Principal Software Engineer at Google specializing in the A2A protocol. He discusses the groundbreaking advancements of OpenAI's latest models, o3 and o4-mini, emphasizing their enhanced reasoning and tool-utilization capabilities. The podcast dives into the rapid evolution in AI, exploring the competitive landscape and innovative breakthroughs like the A2A protocol's role in agent communication. Todd also details security concerns and potential applications, painting a vivid picture of the future of AI technology.

Celebrate a milestone with a deep dive into the AI landscape! Discover the buzz around Meta's Llama 4 launch and the avalanche of announcements from Google Next. Hear about advancements in open source AI, including transformative tools like Git MCP. Explore the performance race among AI models and how collaborative efforts are shaping the future. The hosts also highlight exciting projects aimed at enhancing transparency for developers. Plus, enjoy reflections on a year and a half of incredible AI progress!

Xingyao Wang, Co-founder of All Hands AI, shares insights on the release of OpenHands LM 32B, a groundbreaking large language model fine-tuned for coding. He discusses the significance of open-source collaborations and how this model enhances coding proficiency and application in real-world scenarios. The conversation also touches on recent advancements in AI-assisted tools and the evolving landscape of coding technologies, emphasizing the importance of accessibility in AI development.

Tulsee Doshi, Head of Product for Gemini Models at Google DeepMind, shares insights into the groundbreaking Gemini 2.5 Pro model. The conversation highlights how Google reclaimed the top spot in large language models, emphasizing advancements in context handling and creative capabilities. They explore deep-seeking AI's potential for multimodal interactions, including image and video generation. The discussion also covers innovations in open-source AI models, like DeepSeek V3, and new memory control protocols enhancing AI system interoperability.

Joseph Nelson, CEO of Roboflow, joins the chat to unveil their groundbreaking object detection model and a new benchmark that sets a high bar in visual learning. They dive into the whirlwind of AI advances, including OpenAI's new voice generation technology and innovations revealed at NVIDIA's GTC event. The discussion highlights how these new tools impact developers and the potential for real-time voice applications. Listeners get a glimpse into the fast-paced world of AI where changes happen in the blink of an eye!

Sandra Kublik, Developer Relations at Cohere, dives into the launch of Command-A, a powerful new AI model with impressive multilingual capabilities. She discusses Google's open-source Gemma 3 and how the open-source community is transforming the AI landscape. The conversation covers the exciting advancements in AI technology, highlighting the efficiency of models like EuroBERT and the recent innovations at OpenAI. Sandra also reflects on the podcast's two-year journey and the crucial role of community in its growth.

In this engaging discussion, Junyang Lin, head of Alibaba's Qwen team, introduces the groundbreaking 32B reasoning-focused model, QwQ, and shares its innovative features. Robby Stein, Google's VP of Product for Search, reveals the exciting advancements in their new AI Mode, which enhances user experiences on Google Search. Meanwhile, Dina Kozlov and Jason Kneen dive deep into the Model Context Protocol (MCP), exploring its potential for seamless AI communication and its transformative impact on productivity across various platforms.

In this discussion, guest Kevin Hou, a software engineer from Windsurf, shares insights about their innovative AI-powered coding editor. The conversation dives into the groundbreaking launch of GPT-4.5 and its vast capabilities, including enhanced programming and creative writing. They also explore Claude 3.7's immense potential in coding and the industry's adaptation to rising AI costs. Additionally, the whimsical Grok voice mode adds humor, showcasing how AI is blending innovation with playful interactions. It's a thrilling peek into the future of AI technology!

Leonard Tang, Co-founder at Haize Labs, joins the conversation to discuss their innovative open source evaluation library, Verdict, aimed at improving AI judgment reliability. They dive into the fascinating capabilities of Grok 3, comparing its performance with competitors and addressing censorship challenges. Tang also highlights the impact of nepotism bias in AI models and how Verdict seeks to enhance evaluation efficiency. The podcast explores the exploratory advances in robotics, including robots learning to hand objects to one another, showcasing the exciting future of AI.

Vu Chan, an AI enthusiast and evaluator, shares his insights on exciting AI developments. He recounts a humorous tale of his rogue AI app that turned a simple messaging tool into an unexpected life coach. The discussion dives into the latest open-source releases, including Agentica DeepScaler 1.5B's impressive benchmarks. They also decode the mystery behind OpenAI's confusing roadmap and explore the ongoing innovations that are reshaping the AI landscape. Expect a blend of laughter, curiosity, and thought-provoking advancements!