Send us a text

Major Investments & Structural Shifts

OpenAI and Anthropic now command $92B combined funding, highlighting AI's capital intensity. OpenAI acquired Statsig for experimentation infrastructure. The EU AI Act's reporting rules for large models took effect, while the STREAM standard introduced ChemBio AI safety reporting. Research infrastructure expanded with NVIDIA's Nemotron datasets and the Jupyter Agent Dataset for code execution. Key talent moves include Hugo Larochelle becoming Scientific Director at Mila Québec.

New Tools & Development

On-device intelligence progressed with ChromaSwift for iOS and AI Key hardware. DeepMind released URL Context for processing web content. Open-source advances include trainable World Models, Slime v0.1.0 for reinforcement learning, and Luna-2 for low-latency guardrails. Agent infrastructure improved with xpander's self-hostable runtime.

Model Releases & Research

Notable launches include Apple's FastVLM, Nous Research's Hermes 4 family, Tencent's R-4B VLM, MiniCPM-V4.5, Microsoft's VibeVoice TTS, and Meituan's LongCat-Flash. New benchmarks like AHELM for audio-language evaluation emerged. Training research challenged norms: Transformers excelled without normalization layers using Dynamic Tanh, and "goldfish loss" reduced memorization.

Platform Updates

Perplexity's Comet browser added voice control. LangChain/LangGraph reached 1.0 alpha. Anthropic's code execution now supports bash and persistent containers. Mistral's Le Chat added 20+ connectors and a Memories system. Google's Gemini introduced intuitive image editing.

Global AI Developments

OpenAI plans multi-gigawatt "Stargate" data centers across the U.S., India, and Abu Dhabi. Google expanded Gemini with learning tools, productivity features, and image editing. OpenAI updated ChatGPT with screensharing and video analysis. Walmart launched "Sparky" for shopping assistance, while Amazon debuted Lens Live for product search.

Healthcare & Public Sector

NHS England's stroke AI tool drastically improved treatment times. Switzerland unveiled Apertus, an open multilingual LLM. India launched Adi Vaani to preserve endangered languages. University of Pennsylvania researchers created AMP-Diffusion to generate antibiotic candidates rapidly.

Safety & Workforce Impact

Safety concerns emerged as chatbots sometimes provide risky guidance to suicide queries. Australia proposed banning deepfake apps. Anthropic's CEO warned AI could displace half of entry-level white-collar jobs within five years. Meta is partnering with Google and OpenAI to embed AI across its platforms.

Support the show