OpenAI completes its self-designed AI chip in just 9 months #096

OpenAI has developed a self-designed chip in just 9 months.

OpenAI, in collaboration with Broadcom, has announced an inference-focused chip named "Jalapeño." Aiming to reduce reliance on Nvidia, this chip was developed in an unprecedented nine months and is expected to outperform state-of-the-art chips in performance per watt. With a clear goal of reducing inference costs, there is high anticipation regarding how this will impact the landscape of the tech industry.

Google's model now features screen-operation capabilities.

Google's "Gemini 3.5 Flash" now includes a standard "Computer Use" feature, allowing the AI to directly interact with screens by clicking and entering data on behalf of the user. Processing that previously required specialized models is now possible with this fast model, and integration has already begun in business tools like Salesforce and Xero. During the episode, the host shares an interesting experiment using this feature—be sure to listen to the full episode for details.

AI agents are making waves in the medical field.

"Assort Health," an AI agent that handles hospital appointment bookings, has reached a valuation of $1.2 billion. Its ability to perform not only bookings but also patient intakes and insurance verifications in multiple languages may be evidence that AI is becoming deeply embedded in our daily lives. There is more to discuss, so please tune in to the full episode to hear the rest.