# @max_spero_ — 2024-02-25

♥64 ↻6 · https://x.com/max_spero_/status/1761653804632535331

Google didn't change their image generation system prompt at all from Bard to Gemini. 

It's not laziness, it's an artifact of how Google operates. Any researcher with interest in changing the prompt gets to choose between an interesting problem or a long and arduous task of proving with data that their new prompt is better. 

They will be bombarded with "have you tried [every prompting tip that has ever been written about]" during launch reviews.

They will be asked about benchmarks - how does this affect diversity evals, how does this affect reasoning benchmarks? If even a single number goes down then they now have one specific guy fighting them, the guy whose job it was to make that number go up.

At Google, there's no concept of "good enough." It's perfection or nothing. This works for a lot of things - nobody ships a bad change that hurts search or reduces revenue.

But it's also why so many products are killed - if you're not first you're last. Google had LLM assistants pre-2020 but took years to ship. And it's why people stay in their own lane even when it's obvious that there's low-hanging fruit like the system prompt just sitting there. Not their problem.

tags: author:max_spero_, kind:tweet, on:gemini-1-0, year:2024
cited on: _dossiers/bard-and-gemini-1-0.md, gemini-1-0