What data are you using for those metrics? A 5% improvement in the time to fix bugs or develop features might not be immediately obvious.
All proponents of AI claim they develop 30% faster with AI.
So yes, where are those 30%.
Ask HN: If LLMs are so useful, why haven't we seen any spike in productivity?
11–20 of 58 posts
Re: Ask HN: If LLMs are so useful, why haven't we seen any spike in productivity?
#12Everyone is a decent programmer now who can solve nearly any problem with help from LLM.
Re: Ask HN: If LLMs are so useful, why haven't we seen any spike in productivity?
#13What statistics are you referring to when making these claims?
Github hosts only 20% public repositories. Perhaps open source developers are less likely to have Github Copilot paid out of their own pocket?
Why do you expect "an explosion of new projects" with perhaps 20% of increased productivity? What percentage of open source developers are using LLMs for increased productivity when working on open source? If it's merely 20%, we'd see a 4% increase, something that's hardly noticeable.
Re: Ask HN: If LLMs are so useful, why haven't we seen any spike in productivity?
#14Technology spreads slowly. google docs is an instant 50x productivity increase for any legal process and yet a few years ago I saw an advocate's mind blown by a simple demonstration of simultaneous editing from two people in the same affidavit.
For him, the norm is still to redline a document on paper, and have his secretary add those changes to the original digital document and have that sent over to the opposing team for the same treatment.
I don't have strong opinions about LLMs' coding ability (though compared to the other comments so far I am more on the "LLMs are pretty good at creating software from natural language descriptions" side) but even assuming that LLMs can give programmers a 50x productivity increase, I'd assume it would take 10-50 years for industry and processes to evolve to take advantage of that increase.
Re: Ask HN: If LLMs are so useful, why haven't we seen any spike in productivity?
#15On the B2C side, there's a disconnect because the app developer is typically on the hook for inference costs. This incentivizes them to cut corners so as to minimize the cost of that AI, rather than use it creatively and use high quality models. The apps that let users bring their own keys are much more innovative in this space, but the amount of friction involved in transferring keys keeps most apps from adopting that approach.
Re: Ask HN: If LLMs are so useful, why haven't we seen any spike in productivity?
#16Just wait until we release the next model!
Re: Ask HN: If LLMs are so useful, why haven't we seen any spike in productivity?
#17Youtube, Instagram are flooded with AI generated content.
I believe AI will be useful in Game Dev. AI voice acting, AI face generation. This way all the NPCs will be unique. Possibly AI layout generation.
I don't think using AI to generate script is great use case. It can be used to generate ideas. But still we need human creativity to make great games.
Re: Ask HN: If LLMs are so useful, why haven't we seen any spike in productivity?
#18I actually think all of the things you listed are absolutely happening.
Re: Ask HN: If LLMs are so useful, why haven't we seen any spike in productivity?
#19Just wait until we release the next model!
Do you work at Open AI ?
Re: Ask HN: If LLMs are so useful, why haven't we seen any spike in productivity?
#20They are useful in niches. I got mad at copilot suggesting the wrong thing and went to an ide without it the other day. AST based code completion works much better, copilot gave answers that looked good but were wrong. Worse I needed to hit tab in one place and it kept completing something worng instead of just adding the tab needed there.