Replies: 2 comments
-
Follow-up test results. I tried another test requesting a blog post on a subject. It was a high level overview of Nostr. Since Nostr is relatively new, I figured that Auto-GPT with Google access would be the right tool. I requested a 800 word post. The result was exactly 300 words. I set the parameters to 500 cycles "y -500" after the loading of the tool. It cost exactly $2.00 to generate and took over an hour to run and exactly 261 of the 500. While this is super fun, at this point it seems to be a long way away from being useful. Maybe best way to use is very very specific requests that need recent data. My advice is to keep it short and to the point. Else it gets expensive fast with subpar output. For sure I have not given up on this, but will let it mature a bit more, maybe in another month can try again. |
Beta Was this translation helpful? Give feedback.
-
It seems like it's suffering from perfectionism. Whatever it does is never good enough. I wants to always improve and never move on. Just when you think it has moved on it goes back and does it over. This is THE critical challenge with this software. |
Beta Was this translation helpful? Give feedback.
-
I not sure how to limit the thinking... it seems that if you give it a complex task, for example, "write a blog post on the top 5 ways to cook a steak" -- it keeps refining the results, I have had it continue 100+ prompts and still not finished.
I then tried the simplest task: create an index.html file with the word "hello". In direct ChatGPT it would output this in 1 second. This took over 10 steps as it writes the code, checks it then does something else, etc.
What is the key to limit the thinking, it is a seemingly never-ending self-feeding loop? Any ideas? Maybe someone has found a limiter you can add in 5 parameter statements?
Beta Was this translation helpful? Give feedback.
All reactions