I Spent a Week With Gemini
Pro 1.5—It’s Fantastic

I got access to Gemini Pro 1.5 this week, a new private beta LLM from Google that is significantly better than previous models the company has released. (This is not the same as the publicly available version of Gemini that made headlines for refusing to create pictures of white people. That will be forgotten in a week; this will be relevant for months and years to come.)

Gemini 1.5 Pro read an entire novel and told me in detail about a scene hidden in the middle of it. It read a whole codebase and suggested a place to insert a new feature—with sample code. It even read through all of my highlights on reading app Readwise and selected one for an essay I’m writing.

Somehow, Google figured out how to build an AI model that can comfortably accept up to 1 million tokens with each prompt. For context, you could fit all of Eliezer Yudkowsky’s 1,967-page opus Harry Potter and the Methods of Rationality into every message you send to Gemini. (Why would you want to do this, you ask? For science, of course.)

Gemini Pro 1.5 is a serious achievement for two reasons: 

1) Gemini Pro 1.5’s context window is far bigger than the next closest models. While Gemini Pro 1.5 is comfortably consuming entire works of rationalist doomer fanfiction, GPT-4 Turbo can only accept 32,000 tokens. This isn’t even enough to accept more than a third of Peter Singer’s comparatively slim 354-page volume Animal Liberation, one of the founding texts of the effective altruism movement.  

Last week GPT-4’s context window seemed big; this week—after using Gemini Pro 1.5—it seems like an amount that would curl Derek Zoolander’s hair:

2) Gemini Pro 1.5 can use the whole context window. In my testing, Gemini Pro 1.5 handled huge prompts wonderfully. It’s a big leap forward from current models, whose performance degrades significantly as prompts get bigger. Even though their context windows are smaller, they don’t perform well as prompts approach their size limits. They tend to forget what you said at the beginning of the prompt or miss key information located in the middle. This doesn’t happen with Gemini.These context window improvements are so important because they make the model smarter and easier to work with out of the box. It might be possible to get the same performance from GPT-4, but you’d have to write a lot of extra code in order to do so. I’ll explain why in a moment, but for now you should know: Gemini means you don’t need any of that infrastructure. It just works.

Let’s walk through an example, and then talk about the new use cases that Gemini Pro 1.5 enables. 

Why size matters (when it comes to a context window)