Hacker News Reader

Comment by Insanity | Hacker News Reader

If you think of LLMs as statistically averaging machines and think of human programmer skill as a bell curve, it helps understand the different experiences with LLMs of different people. Those on the left hand side of the skill curve will see an uplift, those on the right hand side will experience degradation.

The more pertinent question in my mind is.. why does it still matter? Your customers don’t care about elegant code, they care about working products. There is some relation to good code and good customer experience e.g in terms of performance, but in general your customers won’t feel the impact from poor code. That assumes both pieces of code are functional though, and you do need to babysit LLMs to make sure it actually remains functional.

View on HN

> If you think of LLMs as statistically averaging machines and think of human programmer skill as a bell curve ...

The problem with this view is that in the scale of LLM progress, this is essentially accurate understanding of archaic models. Modern models are not trained to program using human generated code. They are trained in agentic harnesses where they are able to access real computing environments, and they are rewarded for completing tasks within that environment. This is in contrast the being rewarded for the most likely next token, which used to create essentially an average of human output like you say. Modern training frameworks are able to comfortably produce models that perform well above the center of the bell curve.

> The more pertinent question in my mind is.. why does it still matter? Your customers don’t care about elegant code, they care about working products.

No, they don't care about nice code, but they do care about code that actually works. In my experience, there is a correlation between code that has been thought out correctly and code that hasn't, in terms of security, safety, compliance, data-safety, performance, etc. And so far, all LLM-produced code I've seen is severely lacking.

> That assumes both pieces of code are functional though, and you do need to babysit LLMs to make sure it actually remains functional.

So, I guess we agree on that :)

90% of what even the best of the best programmers do can be accelerated by LLMs. It's not just coding, but research and code archaeology, bouncing ideas around, automated testing etc.

Automated testing is insane. LLMs are amazing at stamping out unit tests, running them, investigating failures, etc.

In fact, I'd argue that [vibe-]coding is the least interesting thing that LLMs can accelerate. I'm really frustrated because LLMs are insanely powerful, but also insanely misused.

On the last point: your customers absolutely feel the impact of your poor code. I’ve worked on very well written codebases and very poorly written ones. The well written ones have had far fewer outages and complaints. When you have multi million line codebases, being “functional” (I understood you to mean this as “working correctly” rather than the technical term) is a spectrum.

Reply on news.ycombinator.com