PySyft VS AIDungeon

PySyft: An open-source library developed by OpenMined that provides tools for building secure, privacy-preserving federated learning systems using PyTorch. Link

https://github.com/OpenMined/PySyft - Federated Learning data science

Incentives are much harder but smart contracts can handle the tech part.

Going this route eventually you quickly have "quantum AI app store" and your system of government is a 12GB download. Can't even say if it's a good idea compared to e.g. anarcho-primitivism.

Something that maybe can help keeping sensor specs secret while still getting critical information out: https://github.com/OpenMined/PySyft

Love this kind of project. To me this is just like https://github.com/open-quantum-safe/oqs-demos/ or https://github.com/OpenMined/PySyft or even k3s so often mentioned in this sub in the sense that I personally don't have a need for it. Yet I find it amazing that us, random curious geeks, have access to this kind of mind blowing technologies for basically free.

Or do I have to actually simulate the whole client server thing because thats how these frameworks do it - Flower and Pysyft .

Some of the technologies which could be integrated include differential privacy, distributed online machine learning, misinformation resilience and multi-party computation, all within the context of smart contracts and bioinformatics.

I think this is how it will be done. Look up PySift for how we can extract high-level insights from private datasets while preserving granular privacy.

What you are talking about is privacy-focused fed ML. Google FLOC is actually trying to achieve something similar. If you are interested in building something for yourself, check this out. https://github.com/OpenMined/PySyft

I have a GPT-Neo AI-Dungeon game in my host. Don't know how to take that to do the same project as yours. But would be interesting. https://github.com/Latitude-Archives/AIDungeon

Its fine-tuning training data is here and it is in the form of choose your own adventures. Programs involving text parsing couldn't be used as training data because it requires the player to type something in.

For what it's worth, the code for the original version of AI Dungeon can be found on the AI Dungeon Github page. If the language stats at the bottom of the page are accurate, then it would seem that, at the very least, the original version of AI Dungeon was primarily written in Python. This is the only info I could find relating to what programming language was used for AID.

Is there a new version of this (or does anyone have a ZIP I could download)? The most recent version here (https://github.com/Latitude-Archives/AIDungeon) is completely broken for python 3.9, tensorflow 2.6, and with my computer's OS (ubuntu 21). It also only includes a .torrent but I have no idea how to make that into the json model.

Its open source, so you can go and look for yourself to see exactly how it was done: https://github.com/Latitude-Archives/AIDungeon

Well, the base GPT-3 model, made by OpenAI, was just scraped text data from all around the internet. GPT-3 was trained on more text than what any person could realistically read in their lifetime. I'm pretty sure that even OpenAI, themselves, don't really know what exactly the AI was trained with. As for Latitude, they finetuned their models on a 30MB dataset, which you can download and look at here. Their finetuning dataset consists of a total of about 90 CYOA games from the website, ChooseYourStory.

Warning NSFW/NSFL AID training material

AI Dungeon's "Censored Words"