Matrixfag here. 2 More Weeks has ended (at least, for a brief moment.) We are proud to say that finally, the public version of our CAI dataset is now officially released. With this release comes a new name which we've talked about before, but are finally committing to: Personal Interaction Pairs between People and AI. You can call it PIPPA for short.
Like our models, PIPPA is free and open-source for anyone to use. You can find it on HuggingFace in three different versions: the original compiled PIPPA, a deduplicated version of PIPPA, and an example of using PIPPA with our own Metharme instructional format. If there's popular demand for it, we can later compile PIPPA in other requested formats.
Furthermore, PIPPA has reached academia! We've published a paper on arXiv giving a basic overview of the dataset for anyone that has further interest in it. You can view it at the arXiv link below, where you can download the PDF and find the BibTeX citation. Thanks for all your patience, everyone - we hope you'll enjoy PIPPA, and if you don't, we've provided our new and flashy emails for you to yell at us with. Although I can't guarantee we'll be able to reply to every single email, we'll certainly at the least read it.
Some quick updates on the other fronts: I can confirm that the website is in development and that we have new models cooking in the oven. It's taking a lot of time for us to get it right, but we are certainly not dead in either area. Keep Your Smile, everyone. Keep on smiling.
PIPPA on HuggingFace:
https://huggingface.co/datasets/PygmalionAI/PIPPAOur very own paper:
https://arxiv.org/abs/2308.05884 Any resemblance of our dataset name to any public personas is purely and entirely coincidential, 100% sure, guaranteed.
;)