Comment by shaky-carrousel
6 months ago
There are open source models, yes, but they are way worse than the commercial ones. I know it because I try every one that comes out. Nothing really compares to commercial ones. And what's going to happen is same story as always. Building a LLM is similar in effort as building a search engine. In the end we'll have one LLM everyone will use. And they'll do whatever they want with it.
Also, what I said is that people using LLMs to write code in programming languages they don't know won't learn to program such languages. Because, as hopefully everyone knows, you learn to program by writing code, not by reading it.
So, you will be a better programmer if you put the effort and properly learn to code in the language, instead of lazily use the LLM clutches. And that's not gatekeeping, that's a fact.
Once you really learn the language, go for it, use the LLM. And then you'll discover the LLM is optimized to find the most average and mediocre solution, the one that takes the least amount of effort to generate. Because obviously they are optimized to be efficient for their owners.
You seem very confident in your predictions. Why would there be “one LLM everyone uses”? That certainly isn’t the case now.
LLMs can help someone learn a new language much faster. If you’re trying to learn anything new at this point and not using LLMs to help, you’re just needlessly slowing yourself down. But that (obviously) doesn’t mean you just ask the LLM to write code for you and commit it without understanding.