I’d like to share my thoughts on the comparison between ChatGPT Plugins and the GPTs models. In my opinion, even though GPTs support multimodality, their logical capabilities don’t quite match up to those of the Plugins.
My observations are based on hands-on testing with both AppCreator Plugin (search on plugin store with name)and AppCreator GPTs(https://chat.openai.com/g/g-dVQNBgzdG-appcreator). During these tests, I specifically focused on their coding capabilities. It became evident that the Plugin significantly outperforms the GPTs in this area. The level of precision, understanding, and execution in writing code was noticeably better with the Plugin.
Given this experience, I sincerely hope OpenAI does not phase out the Plugin, at least not until GPT-5 is released. The unique strengths of the Plugins, especially in tasks requiring higher logical reasoning and complex problem-solving, are invaluable. It’s crucial for the development and progress in AI applications that we continue to have access to such powerful tools.
I’m curious to know if anyone else has had similar experiences or thoughts on this topic. Looking forward to your insights!
Another noteworthy point along this subject is that Plugins will fetch data that is defaulted to hidden in the dropdown (as Forex-Rates plugin can), and that can serve as a reference by expanding and seeing what it retrieved which can be used to verify that it’s output is correct and not a hallucination. For comparison, I haven’t seen any GPTs that can do that.
That’s really a very good function, give us more confidence to the result that might be in hallucination. Plugin can proecess tasks better than GPTs in my experience.
I agree with your observations, but disagree with your suggestion. On the contrary, I hope OpenAI phases out the Plugin immediately, and focuses its energy on GPT to fix its problems, some of which are serious: slow performance, overdone endpoint level security, OpenAPI misunderstanding, etc.
The GPT Store won’t run smoothly with those problems.
The Plugin was an experimental project. It served its purposes well, and revealed its inherent problems in security management and business model. Ditch it now. Beating a dead horse won’t get us anywhere.