I cautiously like this. I asked it to break down “clean the bathroom” and it gave me a reasonable series of steps.
It then gave a time estimate of 3 hours 25 minutes, but I know from my own time tracking that doing a similar series of steps takes me 40 minutes on average.
It seems to overestimate the time taken to do very simple tasks, so if you break the task down too far then it will always wildly overestimate. Perhaps a fix would be to ask the LLM to make one estimate for the overall task, make separate estimates for each subtask, and then ask it to reconcile the two? Something like chain-of-thought/reasoning.
It then gave a time estimate of 3 hours 25 minutes, but I know from my own time tracking that doing a similar series of steps takes me 40 minutes on average.
It seems to overestimate the time taken to do very simple tasks, so if you break the task down too far then it will always wildly overestimate. Perhaps a fix would be to ask the LLM to make one estimate for the overall task, make separate estimates for each subtask, and then ask it to reconcile the two? Something like chain-of-thought/reasoning.