Hacker News Reader

Hacker News

Comment by user43928 | Hacker News Reader

Parent Full thread

user43928

> Making something that's really good takes talent and creativity and AI isn't capable of that

I agree. You need to constantly inject taste and creativity via a lot of testing, feedback, and iteration.

With current technology, this still takes months to years of effort, depending on the scope of the project.

I feel that when you say AI game, it sounds like slop produced near autonomously in a single weekend.

There are other ways to fully use AI for software implementation and asset generation, but still apply standards.

One can spend a full day refining a single gameplay element to tune it into something enjoyable, or to explore and select assets that look great. This can happen with just prompting and testing.

Many skip the effort, and it currently of course results in a bad, barely working product.

View on HN

> I feel that when you say AI game, it sounds like slop produced near autonomously in a single weekend.

That's most of the project I've seen people post about at least. Either actual clones of simple games, or just dreams of saying to AI "Make me red dead redemption but with pirates!" and expecting something playable.

If you spend enough time with AI you likely could come up with a great game. It'd probably take longer to explain to AI what you want, correct what it gave you, explain how you want to teak it, then correct that, etc. than it would just to code it yourself (for those who can code) and as the project grows there's a risk of it becoming unmanageable unless you have AI hyper-compartmentalize everything to fit each function within the limit of what it can keep track of, but that's going to have impacts on performance and maintainability too.

It may not be the most efficient way to do it, but as long as you put in enough time and effort on design, are willing to spend the time forcing the AI to get it right, and don't settle for less than what you envision, it could work.

In my opinion it's an order of magnitude faster with AI.

That things come out perfectly working without need for QA and refinement cycles when writing the code manually seems very unlikely, and the case where coding it yourself is faster than explaining the task to the AI does practically not exist.

> as the project grows there's a risk of it becoming unmanageable

I suspect that what you said there is not based on actual experience but rather something you imagine. In my personal experience AI has shown no issues working efficiently on projects with even 500k LOC.

Reply on news.ycombinator.com