Hacker News Reader

Comment by FuckButtons | Hacker News Reader

You know what computers couldn’t do 10 years ago? Understand and write coherent code. Why do you feel confident in your opinion that they won’t be able to do that in the next 10 years?

View on HN

1) an understanding of what LLMs can actually do

2) a strong history of understanding the limits of functional automation (why are there still project managers? Asana exists!)

3) an understanding that progress will slow in this space considerably once we have made the big wins and the hype cycle dies down. (see also; every tech hype cycle)

I go the exact opposite way on number 3. It seems more likely to me this follows the growth trend of the internet, mobile phones and computer graphics than blockchain and VR. 10 more year of growth like the last would be pretty wild to experience.

"Grog, 10 years ago in 30,000 BC, we didn't even have wheels. We've gone from no wheel, to wheel, in only 10 years. What makes you so sure that self-driving cars aren't just around the corner?"

What we describe as common sense isn't actually common. That is, over-hyped things like microservice architectures saturate the Internet. Since content of the Internet is generally what will become content for LLMs etc it stands to reason they'll be similarly biased towards hype.

Yes, it's certainly possible for content to be curated and models to become biased in other ways but...

If they can "wrangle systemic complexity in a good, sustainable and reliable way" then I think that's AGI. If we hit AGI we'll have plenty of other things to worry about beyond being able to make a living as programmers.

Reply on news.ycombinator.com