Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Who needs Git when you have 1M context windows?

alexmolas.com

Post reply on HN
Print view View original

31–40 of 199 posts

Re: Who needs Git when you have 1M context windows?

#31

Post by the_af »

graemep wrote: post #12
the_af wrote: post #10

I cannot wrap my head around the anecdote that opens the article: > Lately I’ve heard a lot of stories of AI accidentally deleting entire codebases or wiping production databases. I simply... I cannot. Someone let a poorly understood AI connected to prod, and it ignored instructions, deleted the database, and tried to hide it. "I will never use this AI again", says this person, but I think he's not going far enough:…

Its a matter of priorities. Its cheap and fast and there is a chance that it will be OK. Even just OK until I move on. People often make risky choices for those reasons. Not just with IT systems - the crash of 2008 was largely the result of people betting (usually correctly) that the wheels would not fall off until after they had collected a few years of bonuses.

OK, if it's a matter of priorities, let's just ignore all hard learned lessons in software engineering, and vibe-code our way through life, crossing fingers and hoping for the best.

Typing systems? Who needs them, the LLM knows better. Different prod, dev, and staging environments? To hell with them, the LLM knows better. Testing? Nope, the LLM told me everything's sweet.

(I know you're not saying this, I'm just venting my frustration. It's like the software engineering world finally and conclusively decided engineering wasn't necessary at all).

Re: Who needs Git when you have 1M context windows?

#32

Post by the_af »

Aurornis wrote: post #30
the_af wrote: post #10

I cannot wrap my head around the anecdote that opens the article: > Lately I’ve heard a lot of stories of AI accidentally deleting entire codebases or wiping production databases. I simply... I cannot. Someone let a poorly understood AI connected to prod, and it ignored instructions, deleted the database, and tried to hide it. "I will never use this AI again", says this person, but I think he's not going far enough:…

The common story getting shared all over is from a guy named Jason Lemkin. He’s a VC who did a live vibe-coding experiment for a week on Twitter where he wanted to test if he, a non-programmer, could build and run a fake SaaS by himself. The AI agent dropped the “prod” database, but it wasn’t an actual SaaS company or product with customers. The prod database was filled with synthetic data. The entire thing was an ex…

OK, ok, I read the Twitter posts and didn't get the full context that this was an experiment.

I'm actually relieved that nobody (currently) thinks this was a good idea.

You've restored my faith in humanity. For now.

Re: Who needs Git when you have 1M context windows?

#34

Post by cmsj »

SketchySeaBeast wrote: post #7
shallmn wrote: post #5

Earlier quoted context omitted.

I dont know where Gemini stores the context, but if I’m using a local LLM client app, that context is on my machine verbatim.

If you ask the LLM to give you back that context does it give back to you verbatim?

statistically, maybe.

Re: Who needs Git when you have 1M context windows?

#36

Post by Lerc »

big-and-small wrote: post #29

1M context is amazing, but even after 100k tokens Gemini 2.5 Pro is usually incapable of consistently reproducing 300 LOC file without changing something in process. And it actually take a lot of effort to make sure it do not touch files it not suppose to.

With Gemini I have found some weird issues with code gen that are presumably temperature related. Sometimes it will emit large block of code with a single underscore where it should be a dash or some similar very close match that would make sense as a global decision but is triggered for only that one instance.

like code containing the same identifier.

    add_value(thing,other_thing)
    add_value(thing,other_thing) 
    add_value(thing,other_thing) 
    add_value(thing,other_thing)
    add-value(thing,other_thing)
    add_value(thing,other_thing) 
    add_value(thing,other_thing)
Not to mention sneakily functions back in after being told to remove them because they are defined elsewhere. Had a spell where it was reliably a two prompt process for any change, 1) do the actual thing, 2) remove A,B and C which you have reintroduced again.

Re: Who needs Git when you have 1M context windows?

#38

Post by cranium »

dotancohen wrote: post #4

Don't take this as career advice! This is an amusing anecdote. But the only lesson to be learned is to commit early, commit often.

Commit even as a WIP before cleaning up! I don't really like polluting the commit history like that but with some interactive rebase it can be as if the WIP version never existed.

(Side ask to people using Jujutsu: isn't it a use case where jujutsu shines?)

Re: Who needs Git when you have 1M context windows?

#39

Post by this_user »

big-and-small wrote: post #29

1M context is amazing, but even after 100k tokens Gemini 2.5 Pro is usually incapable of consistently reproducing 300 LOC file without changing something in process. And it actually take a lot of effort to make sure it do not touch files it not suppose to.

I have had some very weird issues with Gemini 2.5 Pro where during a longer session it eventually becomes completely confused and starts giving me the response to the previous prompt instead of the current one. I absolutely would not trust it to handle larger amounts of data or information correctly.
Post reply on HN