I want to bring AI Text detection locally instead of relying of SaaS, like quillbot or gptzero. This will be basic to start with
Used copilot autocomplete
I have made readme changed for shipwrights, I also added the performance changes I accidentally forgot to unstash earlier. These just ensure chunks are computed at the same time instead of sequentially. This significantly speeds it up on ARM CPUs because of the low clock speed but many cores.
Log in to leave a comment
I wanted a solution to detecting AI in written text as opposed to buying a subscription to quillbot or GPTZero. This will not be as accurate because they are multimillion dollar companies and I have done this in six - seven hours. There will be false positives & false negatives so does quillbot etc. This site is around the same accuracy as these other sites. There is no other project like this on flavortown, pls (8/9), this was VERY techinical with intergrating AI Models in order to detect AI written text, (9/9), this is very simple to use, there are SOME bugs, (8/9), the storytelling is not that good, I will be the first to say that.
I have now added text highlighting so users can see which words are the most AI. I did this by cross referencing the chunk results and then applying highlighting to the text based on the percentage dynamically. 100% AI makes it 50% strength, 50% AI makes it 25% strong etc. Its all dynamic. I have also changed the model because of accuracy issues with previous models not predicting accurately
Log in to leave a comment
Uh Oh 2 hour 43 minute devlog 
so umm i did a LOT. I have redone the frontend myself so it looks really cool. I made a compose file to make it easier for deploying locally with a file cache. I added a cache for the models so I don’t have to consume all the bandwidth when I am restarting the container lol. I have had some issues with adding darkmode to this, switching back only removed storage, but did not clear the data theme attribute in my code. so yeah. I was trouble shooting my workflow after I migrated from my selfhosted GIT to github. I also adding caching to my dockerfile and my workflow for making and publishing a image.
Log in to leave a comment
Added a basic frontend to this, as well as a CI/CD pipeline to make this into a OCI image for easy deployment. The frontend is currently vibecoded, I am going to pivot towards a human made interface soon since this is all Proof of Concept
Log in to leave a comment
Added POST APIs within phoenix to allow for a basic frontend to be made later on. I have further optimized and improved. idk wat photo since this is backend so yea
Log in to leave a comment
Migrated to elixir, a lot of performance gains, from a couple of seconds down to one for inference via chunking and running in parallel. idk what to do for the photo so idk. enjoy a photo of a bad gateway ig
Log in to leave a comment
I got a working PoC of the AI classifying model working. I have tested my self and the results are very accurate. It took me a while to find a model that was accurate since many are not. This does not work for code just yet. PS idk what to screen shot.
Log in to leave a comment