I feel like some articles tend to downplay LLMs a bit too much. It's not just about form, LLMs do bring some substance as well. When you ask coding questions i

I feel like some articles tend to downplay LLMs a bit too much. It's not just about form, LLMs do bring some substance as well. When you ask coding questions it can often generate code which refers to your unique variable names. It can take what you give it and solve a unique problem which it has never seen before. Like I could give it many functions I just wrote with some names it has never seen before doing some processing it has never seen and it knows how to call those functions correctly and can even assemble them correctly, even though this particular flow is unique to this situation. There is some intelligence involved in rearranging logic and filling logical gaps which goes beyond just rephrasing ideas. I feel like the main issue with LLMs is that they don't have long term memory so it's like talking to an extremely intelligent goldfish. The lack of ability to internalize context is what makes it hard for it to meet goals. You need to keep track of the changing state of reality in order to achieve stuff.