undefined | Better HN

0 pointsunshavedyak9mo ago0 comments

Yea, i'm not even really interested in Gemini atm because last i tried 2.5 Pro it was really difficult to shape behavior. It would be too wordy, or offer too many comments, etc - i couldn't seem to change some base behaviors, get it to focus on just one thing.

Which is surprising because at first i was ready to re-up my Google life. I've been very anti-Google for ages, but at first 2.5 Pro looked so good that i felt it was a huge winner. It just wasn't enjoyable to use because i was often at war with it.

Sonnet/Opus via Claude Code are definitely less intelligent than my early tests of 2.5 Pro, but they're reasonable, listen, stay on task and etc.

I'm sure i'll retry eventually though. Though the subscription complexity with Gemini sounds annoying.

0 comments

sirn9mo ago

I've found that Gemini 2.5 Pro is pretty good at analyzing existing code, but really bad at generating a new code. When I use Gemini with Aider, my session usually went like:

    Me: build a plan to build X
    Gemini: I'll do A, B, and C to achieve X
    Me: that sounds really good, please do
    Gemini: <do A, D, E>
    Me: no, please do B and C.
    Gemini: I apologize. <do A', C, F>
    Me: no! A was already correct, please revert. Also do B and C.
    Gemini: <revert the code to A, D, E>

Whereas Sonnet/Opus on average took me more tries to get it to the implementation plan that I'm satisfied with, but it's so much easier to steer to make it produce the code that I want.

0x4579mo ago

When I use amazon-q for this, I make it write a plan into a markdown file, then I clear context and tell it to read that file and execute that plan phase by phase. This is with Sonnet 4.

Sometimes I also yeet that file to Codex and see which implementation is better. Clear context, read that file again, give it a diff that codex produce and tell it do a review.

ur-whale9mo ago

> It would be too wordy, or offer too many comments

Wholeheartedly agree.

Both when chatting in text mode or when asking it to produce code.

The verbosity of the code is the worse. Comments often longer than the actual code, every nook and cranny of an algorithm unrolled over 100's of lines, most of which unnecessary.

Feels like typical code a mediocre Java developer would produce in the early 2000's

porridgeraisin9mo ago

> Feels like typical code a mediocre Java developer would produce in the early 2000's

So, google's codebase

handfuloflight9mo ago

You were intimate with that?

j / k navigate · click thread line to collapse