AI R&D Is Approaching an Explosive Feedback Loop | Tom Davidson

AI "research taste" is doubling every three months, making a software-only intelligence explosion more plausible than previously assumed.

• TasteVal benchmark: P0 Research's new eval shows AI "research taste" doubling every ~3 months, with models now exceeding average human AI-lab researcher baselines since GPT-5.2/Opus-4.5.
• Narrow but telling: Davidson says the benchmark tests applying off-the-shelf techniques to new metrics, not inventing new architectures—so it's a leading indicator, not proof labs will instantly accelerate.

Themes, sentiment and reasons are AI-written paraphrases, checked so they never quote an episode. They may contain errors. Not investment advice. Episodes remain the property of their creators.