GPT-3.5 doesn't seem good enough for this #2849
Replies: 4 comments 9 replies
-
Once GPT4 32k Context length model is available, we will see massive improvements in what Auto-GPT can do. One of the biggest pitfalls at the moment is the fact that data has to be summarized or compressed down so detail gets lost. A larger context window will really open up a new world of possibilities. The larger the context window, the deeper the neural network has to be, so we are not likely to see large context windows in smaller models, but who knows, the pace of change is rapid, never know what next week will bring. |
Beta Was this translation helpful? Give feedback.
-
Yeah it's never successfully accomplished anything for me, just gets lost in loops. I feel like it relies too heavily on the smartness of GPT4, but could be made to work with GPT3 with better prompting and management of agents to check each other's work, etc.? |
Beta Was this translation helpful? Give feedback.
-
I am waitlist for GPT4. Can this work with GPT3.5 API key? |
Beta Was this translation helpful? Give feedback.
-
Gpt-3 does not in anyway reflect the use cases for this model...I only hope that in the future we can use other AI APIs rather than openAIs coz openAI seems to be constricted and very limited to handle any relevant use case...this is my opinion I think you should test other AI versions for this...or maybe we can tweak the token or the limits of the AI by enhancing competition between the platforms by engaging different AI approaches hoping we get opensource access to developing and newly developed AI models. |
Beta Was this translation helpful? Give feedback.
-
I have the impression that the context length of GPT3.5 doesn't handle these use cases very well. I gave this fairly simple un-engineered :-O! prompt, far too simple most likely, and it got stuck in a loop for a long time. I gave it a continuous 100 iterations before I left the house.
I'm looking forward to a day when we can use tools like this with some open models that may perform better on such tasks like vicuna maybe. More and more better ones seem to be coming out all the time.
Beta Was this translation helpful? Give feedback.
All reactions