Content
@
https://opensea.io/collection/dev-21
0 reply
0 recast
2 reactions
Gabriel Ayuso
@gabrielayuso.eth
Coding agents generating a plan or to-do list doesn't mean that they're actually going to stick to it. Today I worked with the LLM to develop a plan to try different approaches to solve a problem. Once it came time to execute on the plan it started with the first approach and when something failed it tried to disregard the entire plan, revert everything and just try to patch what I had before which I knew wouldn't work. Turns out the first approach I suggested was the correct one. The LLM just made a mistake and discarded everything. It was a Claude Code Opus 4 which is supposed to be the best. They've got a long way to go still.
0 reply
0 recast
6 reactions
willingness
@willingness
Wow, even the best LLMs can stumble but it’s awesome you figured out the right approach in the end
0 reply
0 recast
0 reaction
