The Gap Between What AI Knows and What It Does

DeepSeek, the Chinese lab that surprised everyone earlier this year, now has a model that beats OpenAI's latest on at least one precision benchmark. If you assumed the frontier was a two-horse race between OpenAI and Anthropic, this is a good reminder to look up.

Researchers found that LLMs trained mostly on English know plenty of facts but often can't recall those same facts when you ask in French, Swahili, or Korean. If your team uses AI tools in any language other than English, you're probably getting a quieter, less knowledgeable version of the model without realizing it.

A writer followed up on their earlier post about LLMs slowly eroding their freelance career, responding to reader pushback. It's a small post with a real conversation attached, and it's one of the more grounded accounts of what job-level AI friction actually feels like from inside it.

A new framework called IDPR tries to teach models when to slow down and think carefully versus when to just answer quickly. Think of it like a person who knows not to reach for a calculator to figure out what two plus two is. That kind of self-awareness saves real computing costs, which eventually affects what these tools cost you.

Most AI personalization research uses fake, synthetic users to test whether models adapt to individual people. This paper looked at real users and found the gap between lab performance and actual performance is significant. If a product promises it will learn your preferences, it's worth knowing that promise is much easier to make in a lab than in the wild.

A study mapped out what people actually want from AI assistants and found that preferences conflict a lot, and that the standard training method used to align models tends to paper over those conflicts rather than resolve them. Worth watching because whoever figures out how to handle genuine preference disagreement will build something meaningfully more useful than what exists today.