Hacker News Reader

Comment by midnitewarrior | Hacker News Reader

Do that and the AI might fork the repo, address all the outstanding issues and split your users. The code quality may not be there now, but it will be soon.

View on HN

This is a fantasy that virtually never comes to fruition. The vast majority of forks are dead within weeks when the forkers realize how much effort goes into building and maintaining the project, on top of starting with zero users.

While true, there are projects which surmount these hurdles because the people involved realize how important the project is. Given projects which are important enough, the bots will organize and coordinate. This is how that Anthropic developer got several agents to work in parallel to write a C compiler using Rust, granted he created the coordination framework.

I think the difference now (in case code quality is solved with LLMs) is the cost of effort is now approaching zero.

Good enough AI is not cheap (yet). So at the moment it's more a scenario for people who are rich enough. Though, small projects with little maintenance-burden might be at a risk here.

But thinking about, this might be a new danger to get us into another xz-utils-situation. The big malicious actors have enough money to waste and can scale up the amount of projects they attack and hijack, or even build themselves.

XZ Utils is EXACTLY what came to mind for me.

That exploit / takeover happened precisely because an angry user was bullying a project maintainer, and then a "white knight" came in to save the day and defend the maintainer against the demanding users.

In reality, both the problem and the solution were manufactured by the social engineer, but bullying the maintainer was the vector that this exploited.

What happens when agents are used to do this sort of thing at scale?

This might be true today, but think about it. This is a new scenario, where a giga-brain-sized <insert_role_here> works tirelessly 24/7 improving code. Imagine it starts to fork repos. Imagine it can eventually outpace human contributors, not only on volume (which it already can), but in attention to detail and usefulness of resulting code. Now imagine the forks overtake the original projects. This is not just "Will Smith eating spaghetti", its a real breaking point.

I'm equal parts frightened and amazed.

If your bot is actually capable of doing as your say, why waste time forking OSS repos? Why not instruct it to start 1000 new tech startups and start generating you tons of money? I can "think about" winning the lottery with just as much rigor and effect as day dreaming about the kind of all encompassing intelligence you describe.

Maybe it's time to stop being "frightened and amazed" and come back to reality.

Because everyone's bot will be trying to start a business and make money at the same time, hence nobody will make any money.

Now technically if bots actually improve tech the cost of actual products/services will go down Because of competition.

You keep saying 'come back to reality' but drag someone from 200 years ago to today and our reality would be so shocking they may not recover from it.

> The code quality may not be there now, but it will be soon.

I'm hearing this exact argument since 2002 or so. Even Duke Nukem Forever has been released in this time frame.

I bet even Tesla might solve Autopilot(TM) problems before this becomes a plausible reality.

I mean in 1850 I kept hearing heavier than air flight was just a year away, and yet here we are without heavier than air flight...

I am perfectly willing to take that risk. Hell i'll even throw ten bucks on it while we are here.

Reply on news.ycombinator.com