AI First Pod

Stanford and Google DeepMind unveil a new neuromorphic chip that reduces power consumption for large language model inference by up to 90% while maintaining performance. We also cover: • A new international agreement signed by 28 countries to establish shared standards for AI safety testing and red-teaming • Tesla unveiling a significantly improved version of its Optimus humanoid robot with more n...

Figure AI announces commercial pilots of its Figure 02 humanoid robot in warehouses and factories, backed by a $675 million funding round valuing the company at $6.5 billion.

Hugging Face announces it has now surpassed 1.2 million uploaded AI models, with monthly downloads crossing 500 million for the first time. We also cover: • Meta releasing Llama 4 Maverick, a powerful new 405B open-weight model already topping several community leaderboards • The European Union proposing mandatory transparency reports for all AI systems above a certain capability threshold • Perpl...

Hugging Face announces it has now surpassed 1.2 million uploaded AI models, with monthly downloads crossing 500 million for the first time. We also cover: • Meta releasing Llama 4 Maverick, a powerful new 405B open-weight model already topping several community leaderboards • The European Union proposing mandatory transparency reports for all AI systems above a certain capability threshold • Perpl...

OpenAI releases o3-pro to all ChatGPT users (Free through Enterprise), bringing frontier-level reasoning in science, math, coding, and complex multi-step tasks with significantly improved efficiency over o1.

NVIDIA launches the Vera CPU—purpose-built for agentic AI with twice the efficiency and 50% faster performance than traditional CPUs for data processing, training, and inference at scale.

NVIDIA launches the Vera CPU—purpose-built for agentic AI with twice the efficiency and 50% faster performance than traditional CPUs for data processing, training, and inference at scale.

 Mistral releasing Mixtral 8x22B v0.4 with significant gains in multilingual reasoning, instruction-following across 70+ languages, and reduced hallucinations. Scale AI closing a $2.5 billion enterprise partnership with a major global bank to build custom, auditable agentic workflows for compliance, risk, and reporting.  

NVIDIA launches the Vera CPU - purpose-built for agentic AI workloads with twice the efficiency and 50% faster performance than traditional CPUs for data processing, training, and inference at scale.

NVIDIA launches the Vera CPU - purpose-built for agentic AI with twice the efficiency and 50% faster performance than traditional CPUs for data processing, training, and inference at scale.  

xAI launches Grok-3.5 for X Premium+ subscribers - a reasoning powerhouse that leads or ties frontier models on hard benchmarks with lower latency and strong tool use.

NVIDIA launches the Vera CPU - purpose-built for agentic AI with twice the efficiency and 50% faster performance than traditional CPUs for data processing, training, and inference at scale.  

xAI launches Grok-3.5 for X Premium+ subscribers, a reasoning powerhouse that leads or ties frontier models on hard benchmarks with lower latency and strong tool use.  

xAI launches Grok-3.5 for X Premium+ subscribers, a reasoning powerhouse that leads or ties frontier models on hard benchmarks with lower latency and strong tool use.  

ChatGPT saw a 295% day-over-day spike in US app uninstalls after news of OpenAI's Pentagon deal broke, with many users citing discomfort with the military partnership, while Anthropic's Claude briefly topped the US App Store free charts as an alternative.

xAI launches Grok-3.5 for X Premium+ subscribers, a reasoning powerhouse that leads or ties frontier models on hard benchmarks with lower inference latency and strong tool use.  

NVIDIA launches Cosmos, an AI-powered simulation platform that creates photorealistic, physics-accurate virtual worlds for testing robotics, autonomous vehicles, and industrial systems at 100× the speed of traditional tools.

The rise of creative AI agents. Startup Luma AI has launched a system that can generate entire campaigns across text, images, video, and audio from a single prompt, pointing to a future where AI manages full creative workflows.

OpenAI rolls out o3-mini to all ChatGPT users (Free to Enterprise), delivering near-frontier reasoning in science, math, and coding with 60–70% less inference compute than o1-preview.  

OpenAI internally rolls out 'Project Strawberry'- a reasoning-focused model family hitting near-human performance on hard science/math benchmarks with 5–10× less inference compute than frontier models.  

xAI launches Grok-3 for X Premium+ subscribers - a reasoning powerhouse that competes with GPT-4o and Claude 3.5 Sonnet on hard benchmarks while using less compute.

Yann LeCun publishes a new paper arguing AGI is misdefined and proposes "Superhuman Adaptable Intelligence" (SAI) as a clearer goal - systems that exceed humans in rapid adaptation, few-shot learning, and open-ended generalization.

The US Department of Defense has officially designated Anthropic a "supply-chain risk" under the Defense Production Act after the company refused to remove safety restrictions on Claude for military use - prompting CEO Dario Amodei to call the label "baseless" and vow a court challenge.  

 The term "agentic slop" is spreading fast- industry voices at MWC and in VC circles are calling out flashy but shallow agent demos that impress in 30-second clips but fail in real-world complexity, signaling growing hype fatigue just as agentic AI is supposed to scale.  

Venture capitalists are drawing a hard line: no more funding for "AI wrappers" or hype-driven SaaS startups - investors now demand deep moats, proprietary data, and real unit economics over generic LLM interfaces.