ChatGPT For Fault Finding - Amazing Solution

[Anthony Parsons]

Maybe something many haven't tried, but I had an issue with Elasticsearch stopping after x period, basically, resetting itself and clearing the database, so only newly posted content was being shown in histories and search. I gave ChatGPT a go to see if it could find the problem - and well, and hour later after it literally walked me through everything, including all command line instructions, copying and pasting back logs and such, it found the issue for me, which server techs couldn't find already.

The server is Almalinux 9 with WHM, and it was running Imunify360 which was causing Elasticsearch to stop when its cron ran, causing a conflict and thus the issue with ES and the forum.

ChatGPT explained everything, started in one place, ended up in various other server logs and from all the pasted log entries for analysis, found the issue and told me how to resolve it.

Just want to remind people of this easy to use service that is pretty freely available nowadays, which can solve your tech issues pretty damn quick.

 

[eva2000]

Yes, very useful. But I wouldn't just rely on one LLM model/AI service for the answer, as it can hallucinate. Also, even the same LLM model from different providers/implementations can result in different results. I'd consult more than one LLM mode/AI service and have them evaluate each other's responses for accuracy and hallucinations. As such I built in an eval mode for my AI tool so multiple LLM models can evaluate each others response to the same prompt/question or-cli.py multi-model Evaluation

I usually use the Sublime Text editor and I prefer it over Visual Studio Code. So a few weeks ago, I revisited Visual Studio Code + Github Copilot Pro + Claude 3.7 Sonnet Thinking for one of my scripts to see how it compares to my usual direct Claude Pro/ChatGPT Plus usage and it failed badly, giving me 8/10 inaccurate recommendations.

I asked Copilot Pro/Claude 3.7 to comment on GPT o3-mini high's evaluation of Copilot's response, and this is what it gave me - admitting it got the majority of its recommendations wrong!. Seems the issue is with Github Copilot Pro's Claude 3.7 Sonnet as direct with Claude AI Pro it got an accurate response.

 

[Suzanne O]

No thanks...