Quoc Le says Gemini 3 reasons well from internal knowledge alone
Original titleI've also been using Gemini 3 for fun and its ability to reason with internal knowledge is so good. For example, I gave it the prompt "Vi...
AISummary
Quoc Le reports that Gemini 3 autonomously identified the components of Neural Architecture Search with Reinforcement Learning, wrote p5.js code to animate it, and explained the concept clearly from a single prompt. He presents this as an informal example of the model's reasoning from internal knowledge, not a formal benchmark.
Source: Quoc Le · x.comPublished · added here