AI Roundup: Gemini 1.5, Sora, ChatGPT Memory, Karpathy Exit, Altman's Chip Fund

The AI world is moving fast this week with several major announcements. Google rolled out an upgraded Gemini 1.5, significantly expanding its context window to 1M tokens, allowing the model to process hours of video or massive codebases in a single prompt. OpenAI also pushed updates to Sora, its text-to-video model, improving video coherence and length, while ChatGPT began rolling out a persistent memory feature that lets the assistant recall past conversations and user preferences across sessions. These upgrades point to a trend toward more personalized and context-aware AI systems.

In talent and infrastructure news, Andrej Karpathy, a founding member of OpenAI and former Tesla AI director, announced he is leaving the company again, citing a desire to focus on research and education. Meanwhile, Sam Altman is reportedly in talks to secure trillions of dollars in funding for a massive AI chip manufacturing initiative to address GPU shortages. Stability AI also released Stable Cascade, a new text-to-image model built on the Würstchen architecture, delivering higher quality images at a fraction of the compute cost of Stable Diffusion. Together, these developments highlight the industry's rapid push toward larger models, richer multimodal capabilities, and the hardware required to power them.