You are viewing a single comment's thread from:

RE: Further analytics regarding comments on HIVE

in Hive Statistics22 days ago (edited)

Yes make a ranking of the most swearing users!

How you determine if a user is human or not? You check manually some comments or? Not sure if you have already or how hard would it be to integrate an Ai checker, quite some comments are obvious ai

I guess @hive-197333 snaps work gave me a good boost on there

Sort:  

I have read the duplicate comments of the user, and manually determined whether they're human, or just a curation service :P

In terms of AI checker, that will be something for a later revision, I will need to find a way to process the comments on a local LLM for the "confidence" that it is AI generated, as I cannot afford an API end point to check each comment, the computational costs would be astronomical, but if I can do it on my local GPU - then I will :P (When the sun is shining and I have solar power to do it for free!)

So I think my definition of "Human" is "Someone who doesn't make the same comment multiple times, constantly, and isn't on the exclusion list that @friendlymoose gave me that is used for the @topcomment project.

There's still much work to be done on this analysis, in order to make it simple and easy to digest for everyone on HIVE, just not nerdy analysts like myself.

A local llm could be good to go, I'm unsure how "heavy" it is, I only used stable diffusion locally... Worth a try tough

A first version can never be complete but it's pretty clear, you can keep elaborate over time

I have a 4090, it should be no problem. Have also run SD, Flux, and other models locally in the past. It is fun to play with. Comfy UI is a bit of a nightmare though. LM Studio is pretty easy to use for LLMs, though.

Yep, I like the local as you can tweak and add lora and such how you prefer rather than be forced to use what is offered online... Oh comfy, I never tried it, it looked complicated but people showed good results with it

I used all of the three major front ends for image gen locally, Automatic1111, Invoke, DrawThings on ipad / MacOS, but Comfy UI is ... the most flexible, and I think has the most extensibility.

You do tend to end up with spaghetti string node combinations, reminiscent! of any node based software, which can often end up looking like...

🤣

I only used A1111, my pc is not really new, it has some years it takes like 20 minutes per image considering I added ADetailer too... I tried forge, it's way faster but it's buggy for me with some models and lora