This Week in AI: The Cost of Progress

Google and Mistral announced new frontier AI models this week as investors grew more cautious about the billions flowing into AI development. On the latest episode of This Week in AI, host Christina Stathopoulos, founder of Dare to Data, discusses new releases from Google, Mistral, and DeepMind, along with growing concerns about the financial returns on AI investment. Anthropic's IPO plans, OpenAI's latest funding round, and a tech stock sell-off suggest investors are paying closer attention to how much AI development costs and when those investments might pay off.

Researchers are testing new approaches to understanding and improving AI models. Goodfire wants to map how language models process information, while Google is investigating whether a "sleep" phase could help models retain knowledge. Advances in diffusion language models offer another possible route to faster text generation. The responsible AI discussion addresses some less visible consequences of adoption, including the "tokenizer tax" that makes AI more expensive in certain languages, Google's data center investigation in Finland, and a proposed standard that would let websites restrict the use of their content for AI training.