Is it time to implement Ai?

[chrsbe]

An AI tool with full access to the forum posts could be very helpful in bringing together fragmented information in a compact way. I have no idea how much work the implementation would be, but maybe there’s already a ready-made solution?

 

[emax]

I've seen quite some AI "solutions" giving me absolute wrong information and stupid answers. May be in the future it will get better. But so far, none of it is IMO really usable. It's mostly just hype.

But you can try on your own to find out how good or bad chatGPT works. If you enter chatGPT.com, just type this text ...

Can you analyse theineosforum.com postings to find a solution for how to fix too warm air from the grenadiers air condition?

... and chatGPT will give you an answer.

You can try it and tell us how good or bad chatGPT's answers were.

But it may be better to ask an own (different) question because I already "contaminated" chatGPT with my idea.

 

[chrsbe]

If you get so many wrong answers that chatgpt is unusable for you try refining your prompt. You should restrict search to the ineos forum only and request links to the answers. Not very convinient but it works. Altough Chatgpt will tell you that it can't access all informations. Having an Agent integrated in the search here in the forum would be on another usability level.

 

[theox]

If there was a way to export all the text from all the posts in this forum, I would def try to load it into a NotebookLLM and see what happens.

 

[Clark Kent]

If you get so many wrong answers that chatgpt is unusable for you try refining your prompt. You should restrict search to the ineos forum only and request links to the answers. Not very convinient but it works. Altough Chatgpt will tell you that it can't access all informations. Having an Agent integrated in the search here in the forum would be on another usability level.

Agreed. Like all searches the SISO principle applies. in equals out. What I do like about AI searches is you can refine the search as a running dialogue instead of starting over again.

You: Create a picture of a unicorn.
ChatGPT: Ok here's your unicorn.
You: Make it a purple.
ChatGPT: Ok here's your purple unicorn.
You: Put a saddle on it.
ChatGPT: Ok here's your purple unicorn with a saddle.

Google can be forced to search a single website by adding site: website after the search term. Example:

nato socket site: theineosforum.com

And you can now do a Google AI search on the same website by selecting AI mode at the top of the page.

 

[emax]

You should restrict search to the ineos forum only

It don't know whether you read my posting. That was what exactly what I did.

Concerning the "prompts" (I guess you mean my inputs): I don't think my inputs are too bad for something which calls itself an "intelligence". Wrong answers were not wrong in that my questions were unclear. The answers are too often factually incorrect, like "what is 3+4" would get an answer like "12.5". And it is not just chatGPT. There are many so called AI websites which deliver ridiculously bad results.

But many people don't even take note of such errors.

I only use AI occasionally for research tasks that are difficult to formulate – or when I simply don't know how to search a complicated matter. But in my opinion, that has nothing to do with "intelligence," but rather with good access to comprehensive datasets and a well-implemented search function.

I once asked for a C-implementation of the Dijkstra algorithm, a well known method to calculate a route in a mesh and not at all rocket science. The result was catastrophic. I asked again and pointed chatGPT to the errors: No way, other errors popped up, chatGPT was not able to deliver a syntactically and semantically correct result. It was impossible to compile. That was not a matter of "asking wrong", because obviously chatGPT understood very well what I asked for. (Today, you would get a (at least syntactically) correct result, even in C++ which once yielded a total mess when I asked for it).

Just an example. But If you want an AI here, no problem for me, to each their own. But I don't know whether I would trust it.


BTW: chatGPT for a forum is not for free ...