Beyond Memorization: Do Larger Models Know More, or Just Better?

via

OpenAI

Just read 2 papers: 1. Incompressible Knowledge Probes 2. Densing Law of LLMs densing laws suggest for every 3 months you will get a new model that does same things in half the parameter. These ppl (IKPs) argue that better architecture and training methods only affect instruction following, reasoning abilities and stuff but not factual knowledge(this is still parameter dependent and scaling rules still apply). This kinda leads to the question like how much factual info is enough? like we can do

More from OpenAI

A Dark-Money Campaign Is Paying Influencers to Frame Chinese AI as a Threat

OpenAI·6

Stop bloating your agent context with MEMORY.md. I built a local cognitive memory MCP instead.

OpenAI·5

Cant do biology any more (nudity, gore) can't even do a local security scan, I wont be using cloud services much longer

OpenAI·7

OpenAI Has Just the Right Person to Curb Its Wild Spending

OpenAI·4

How it feels to do biotech in 2026

OpenAI·4

4 CLI cmds and your AI knows more about your project than you do (open source)

OpenAI·4