Ask HN: Are we gonna back less powerful local LLMs
As developers, We are using very powerful llms.But pricing are getting higher day by day.Many developers start to get mac mini for local llms.At the end of the day, are we gonna back to not very powerfull llms because of costs? What Do you think?
16 comments
I think the trend is that top models are meant for companies. Small devs did our job of hyping and training and we can now either pay way more, or pay more and use not sota models, or give our data to access train chinese models in hopes they keep 6 months behind in the cold war and need still need some of our input, or invest around 5-10K for powerful local personal AI [0].
On the other hand I think that AI can really raise the bar of "average tech", and we devs are wired to think that better tech == more value... but this might not be the case in the many many many cases where existing average tech and velocity is already good enough and the real moat is the handshake, trust, marketing, etc etc
[0] https://github.com/antirez/ds4
On the other hand I think that AI can really raise the bar of "average tech", and we devs are wired to think that better tech == more value... but this might not be the case in the many many many cases where existing average tech and velocity is already good enough and the real moat is the handshake, trust, marketing, etc etc
[0] https://github.com/antirez/ds4
I saw a tweet that eventually companies will realize it's more expensive using AI that they then return to human developers. Lol. The irony that would be.
Funny I just came across a comment that accurately answers your question: https://news.ycombinator.com/item?id=48047722
This is one take. Not every project is made better by expanding the scope by 100x. This often makes things worse.
One of the hardest things to do in software is to make things simple and easy to use. That could mean increasing the scope to do so. Scope doesn't automatically mean more bloat.
Thanks!
I need a proficinal hiking system