Although OpenAI says that it doesn’t plan to use Google TPUs for now, the tests themselves signal concerns about inference costs.| Network World
The new chip is designed to run LLMs that support reasoning, which typically require more compute to generate each response.| Network World