Hacker News Reader

Comment by tmountain | Hacker News Reader

I’ve been writing software for 25 years and am “vibe coding” a fairly complex game in my spare time. It saves time on boilerplate but it is absolutely not capable of doing all the work. There’s still some assembly required and doing so requires domain knowledge and expertise. Also, prompting properly requires the same. If I were to compare the experience to building a house, I’m now the foreman—and no longer doing the manual labor. There’s a real risk of systems making it to production that nobody understands, but we have that today.

View on HN

There’s no such thing as vibe composing music, even though experimenting with knobs is not writing notes, and even though writing notes seems like writing programs.

À music score may be complete since very first attempts at it, while arrangement and sound désign may be added later. Writing actually catchy music is much more difficult than writing a todo app even though they may seem similar in engineering complexity.

Coding is not composing and vice versa. Code which produces music scores is not what audio models do.

My comment wasn’t in relation to music. It was a response to the parent comment about vibe coding.

The scary thing is while a construction labourer knows how to properly lay bricks, an AI's output is reliably unreliable. Does the foreman have to go round checking every little thing on the site? That would be a waste of time. Using these AI tools for anything important is too high risk.

> Does the foreman have to go round checking every little thing on the site?

Uh, is that not half of the foreman’s job? They’re there to direct and coordinate the work, resolve unforeseen issues, and to enforce the required quality of work.

I'm certain they don't check every brick. In the same way I shouldn't have to check simple maths. Today's "AI" can't even reliably add two numbers, it's ridiculous.

There exist many problems for which it's easy to verify the inputs-outputs, but much harder to write the functions to convert the inputs into the outputs.

Just write an executable spec and have the AI generate the code that fulfills it. Where is the risk?

I guess the risk would be that the cost of compute would explode out of control long before you got a viable solution.

To give a bit of an absurd example, I imagine there's a reason it's not currently worth it to generate the code with a random character generator, even though hypothetically that would get you there eventually. If we consider AI a much, MUCH better version of a random character generator (let's say it's a million times faster. No, let's say that it'll get you the solution in quadratic time instead of factorial time), that doesn't necessarily mean it's actually worth it now.

Just time box it then?

Ok in that case I guess the risk would be that it simply wouldn't work and you'd lose whatever time and compute happened in the time box. Which would be a pretty cheap price to pay, but I imagine you'd only try it a few times before giving up and using a different strategy.

Reply on news.ycombinator.com