The @financialtimes talks about how GPT-5’s middling performance on benchmarks may be a sign that the approach of scaling LLMs by training them with more data and compute has reached diminishing returns.

The article also argues that the Trump administration being comfortable with Nvidia selling GPUs to China is this sense that AI isn’t getting exponentially smarter nor is a national security threat.

Even if LLMs have plateaued, AI apps are just getting started.

https://www.ft.com/content/d01290c9-cc92-4c1f-bd70-ac332cd40f94