why doesn’t openai use chatgpt for answering document related questions (via a chatbot) on its website?
1 Like
Hi!
But you can use chatgpt for exactly that purpose! sometimes you need to hint that it should use bing to get the most up to date information. What problems are you having?
'… Im afraid i have to aggressively |disagree| with this… ’
Okay, I’ll use ChatGPT for exactly that purpose:
Me:
Why doesn’t OpenAI use ChatGPT technology for answering product documentation-related questions, by placing an application-specific chatbot on OpenAI’s website? Is the reason because most people are not yet fluent in phrasing their questions correctly for an AI?
ChatGPT:
There could be several reasons why OpenAI might not have chosen to deploy ChatGPT directly for product documentation-related questions on their website. One key factor might indeed be the challenge of users phrasing their queries in a way that the AI can effectively understand and provide accurate answers. AI systems like ChatGPT work best with well-formulated and clear questions, and sometimes technical or domain-specific inquiries might require a more tailored approach.
Another aspect might involve the complexity and specificity of product-related information. While ChatGPT can generate responses based on its training data, product documentation often requires precise and up-to-date details that might be better handled by a dedicated system specifically designed for that purpose.
Additionally, there could be considerations around security, privacy, and the need for fine-tuned control over the information provided through a chatbot on their website. OpenAI might be exploring or developing more specialized AI systems or tools better suited for such use cases that demand accuracy, security, and domain-specific knowledge.
Creating an application-specific chatbot involves a deep understanding of the domain and targeted user needs, and it’s possible that OpenAI is exploring ways to harness the capabilities of AI, including ChatGPT, in more tailored and specialized applications for handling product documentation-related queries.
The truth is the AI doesn’t have the quality to turn documentation-based questions such as “how can I extract the token usage from a chat completion in Python” to “OpenAI decided to decode RESTful API json into their own Pydantic objects, making model.usage a custom CompletionUsage object, that while appearing dictionary-like, is not subscriptable. You must use a Pydantic model object method such as .model_dump() or a custom method such as .total_tokens int object on the CompletionUsage object”, no matter how much you fill the AI with context (and for that, little documentation).
1 Like
I was gonna say works on my machine, but I don’t use the libs 
I don’t know what value they add. But if it’s true that they’re trash then no Chatbot is going to answer your question to satisfaction
Babbage comes to mind
On two occasions I have been asked, 'Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out? ’ I am not able rightly to apprehend the kind of confusion of ideas that could provoke such a question.