Hard Fork says top AI models still lead, but others are catching up

In short: The New York Times podcast Hard Fork says the biggest AI models still perform better overall, but the gap is shrinking in certain tasks.

What's going on

Hard Fork is a technology and AI podcast from The New York Times, hosted by journalists Kevin Roose and Casey Newton. In a recent segment, Newton pointed to a clear performance gap between the largest so called “frontier” AI models and most other models.

“Frontier” models are the very large systems made by companies like OpenAI, Anthropic, and Google. You can think of them like top of the line cars that cost more to build and test. Many other models, including open source models (ones people can download and modify), smaller “distilled” models (compressed versions designed to be cheaper and faster), and many Chinese commercial models, often do not match them on the hardest tests.

At the same time, Hard Fork also highlighted signs that the gap is narrowing in specific abilities. An episode description in the show’s public feed references “Chinese A.I. Models Close the Gap With Anthropic and OpenAI,” which signals that the hosts see meaningful progress from Chinese AI companies. The key idea is that even if the best models still lead overall, competitors may match them on some tasks, like certain writing, coding, or question answering tests.

What to watch

Watch how quickly these “good enough” models spread, especially if they are cheaper to run or easier to access. If more models reach similar quality for everyday tasks, businesses and governments may have more choices, and rules about safety and access may become harder to apply evenly.

Source: NYTimes