The Limitations of LLMs
As much as I love the new technology from Anthropic, it still has the same limitations that all LLMs have had since ChatGPT was released.
They eventually forget what you're doing and what they've done for you.
There needs to be two things integrated into LLMs to make them effective.
- The ability for me to mark code as blessed and not to change it going forward.
- Remember every conversation from front to back. If this means invisibly rerunning every prompt before running the current prompt, so be it. I'd give up a few seconds of wait time to keep moving forward.
I know a lot of people are trying to do both of these things. Until it's implemented, LLMs are only good for small things.