P-Zero Research

[P-Zero Research @pzeroresearch]

@deanwball We just published our empirical research into this question: https://t.co/8I47UHV5pc

In the context of AI R&D, we operationalise experimental taste by how effectively a given amount of compute can be utilised.

[P-Zero Research @pzeroresearch]

How fast is AI's research taste improving?

We find that the experimental research taste of frontier models has doubled every ~3 months since December 2025. The best model, Opus 5.5, now exceeds our expert human baseline. Our human experts are experienced researchers, but most haven’t worked at a frontier lab.

Why measure research taste? In the AI Futures Model, it largely determines how quickly artificial superintelligence is reached once coding is fully automated.

[P-Zero Research @pzeroresearch]

We're hiring! Check out our available roles here: https://t.co/bOpPuqJqgp

Read more about TasteVal: https://t.co/qLzIAnQcbN

[P-Zero Research @pzeroresearch]

How fast is AI's research taste improving?

We find that the experimental research taste of frontier models has doubled every ~3 months since December 2025. The best model, Opus 5.5, now exceeds our expert human baseline. Our human experts are experienced researchers, but most haven’t worked at a frontier lab.

Why measure research taste? In the AI Futures Model, it largely determines how quickly artificial superintelligence is reached once coding is fully automated.

[P-Zero Research @pzeroresearch]

TasteVal's tasks could help labs hill-climb on research taste if released, so we're publishing the results and methodology but keeping the tasks private. Here's our reasoning, and what would change our minds.

https://t.co/fvtUOtwQEL