I mean, I just the last hour setup the deepseek-coder-6.7B-instruct-GGUF model on my computer and tried it out with llama.cpp. It is maybe free ChatGTP level.
If I could have run this when I was a student it would have been a real killer app. It is kinda impressive that I can run it on a crappy laptop.
I some way this is also really scary, job wise. I would not recommend anyone getting into the programmer fields after the last years progress in these tools.
But at least they will not be locked up to "OpenAI"'s servers ...
I wonder if there could be a community effort doing model training, like the protein folding? If 30 000 computers ran some model training while not being used, would that be enough to get anywhere?
Not getting into programming because anyone can use an LLM to do programming does not suddenly fill the productivity needs in an absolute sense. People can do more with less which also means we will quickly raise the demand for the new things we need to do.
I think it will flip the tables completely. When "we" got into programming you more or less had to have aptitude for it and wade through dull documentation and still think it was fun to get anywhere. And there was a geek stigma. The recruiting body was really small for future programmers.
LLM are so much help to get a newbie to some sort of entry level it gonna flood the market compared to when we entered. And the pay has got really good too, in itself a big enough problem leading to flooding from bellow.
That principle of demand increases up to the size of the supply makes sense until the supply becomes effectively infinity, which is what AI is doing. I'm not sure demand rises infinitely. I do believe that programming skills are heavily devalued in a world of effective AI, and most people on HN are a bit delusional about this fact, mainly because it's depressing to us computer touchers...
If I could have run this when I was a student it would have been a real killer app. It is kinda impressive that I can run it on a crappy laptop.
I some way this is also really scary, job wise. I would not recommend anyone getting into the programmer fields after the last years progress in these tools.
But at least they will not be locked up to "OpenAI"'s servers ...
I wonder if there could be a community effort doing model training, like the protein folding? If 30 000 computers ran some model training while not being used, would that be enough to get anywhere?