You might think that writing a blog article must be easy with LLM, but I’ve just deleted many sections of generated text. The more your engineering expertise grows the more code you remove. Less is better.
Data > LLM #
Everyone talks about benefits of LLM but people don’t aware that the LLM needs sources of truth. And the reason why LLM “knows” the answer to a question is because there is an article or a book or a post somewhere on the internet written by human. So what will happen when articles become artificial? The responses of LLM will be less correct. The amount of sharing the real knowledge will decrease. Look at the StackOverflow after ChatGPT launch, how many users stopped reading answers from real humans. The most important issue right now is that LLMs are breaking a feedback loop.
Why LLM generated code is wrong and developers still have to open StackOverflow? #
If an answer from stackoverflow is wrong you can click the ArrowDown button to mark that answer as not useful. And the most important is the ability to propose back to stackoverflow a new correct answer adjusted to today’s reality.
Do you write a correct answer back to LLM? no… #
How people should tell correct answer back to LLM? Shouldn’t we do that? If people start writing wrong answers to stackoverflow their answers will be quickly down-voted. Can you down-vote a training data for LLM?
Let me know your thoughts in the comments below.
