Very interesting observation. However, if an LLM has access to an IDE, memory, CoT, etc., why could there not be multiple instances of the LLM running and debugging code in parallel to develop more well-rounded understanding of the language and write documentation? A consequence could be an increase in the number of possible programming languages and more rapid adoption with tools like Copilot. Another consequence could be whole new programming paradigms as AI becomes more sophisticated, including more intuitive ways to map human language onto the digital world.
I think we'll definitely come up with clever ways to create at least some of the data needed to train an LLM on a new programming language, but I think it will be difficult to quickly integrate this into large LLMs
Thank you for sharing the insight.
You're welcome glad you enjoyed it
Auto-plagiarize needs input. No input, no “intelligence”.
Very interesting observation. However, if an LLM has access to an IDE, memory, CoT, etc., why could there not be multiple instances of the LLM running and debugging code in parallel to develop more well-rounded understanding of the language and write documentation? A consequence could be an increase in the number of possible programming languages and more rapid adoption with tools like Copilot. Another consequence could be whole new programming paradigms as AI becomes more sophisticated, including more intuitive ways to map human language onto the digital world.
I think we'll definitely come up with clever ways to create at least some of the data needed to train an LLM on a new programming language, but I think it will be difficult to quickly integrate this into large LLMs
Yes, I was strugling with LLM cannot do Gleam
not surprising, it seems like a very cool language but I hadn't heard of it before
aliens laughing at us for getting stuck with JS bs forever!
They'll need to invent JS to make it to earth in the first place
@ nah, nasa didn't have JS when they sent voyagers into deep space