This is helpful as we think about how schools with different regulations and cultures can adapt, align, and audit the AI they use for education. Students and teachers will be the primary individual users, but there are many stakeholders with a voice in shaping its behavior.
First off, huge thanks to Audrey for the thoughtful reply she sent me about KAOS. The points about critical mass and spam risks are dead-on and we’re already using them.
For anyone curious: KAOS is nothing more than a giant, delete-proof, global filing cabinet for human opinions. Raw data sits there forever, untouched. Every single tool, app, or view that people actually look at gets to filter it however they want (your GAC method, block lists, reputation, whatever). The filing cabinet itself never gets cleaned or sorted by us.
We think the first killer use is replacing every crooked review site on earth (restaurants, doctors, politicians, you name it) with one place that can’t bury or delete what you say.
Would love to hear any ideas from this crowd on solving the chicken-and-egg adoption problem without ever touching the raw data.
Really interesting work, thanks for pushing forward with these ideas. Have you also considered that communities might want to "reverse" some of the learning of an AI? For example, a community is really happy with their AI but collectively feels some small changes are needed, but having made them the model is now worse. Is there a need to be able to revert to a previous state? This might allow for de-risking prototyping new input and exploring alternatives.
This is helpful as we think about how schools with different regulations and cultures can adapt, align, and audit the AI they use for education. Students and teachers will be the primary individual users, but there are many stakeholders with a voice in shaping its behavior.
Here is the general idea: https://home.xtramath.org/blog/xtramath-ai-building-public-infrastructure-for-universal-math-fluency
Thanks so much for this. Your work is very inspiring. We would love to help contribute to this and will take a look at the github.
Audrey & the CIP folks,
First off, huge thanks to Audrey for the thoughtful reply she sent me about KAOS. The points about critical mass and spam risks are dead-on and we’re already using them.
For anyone curious: KAOS is nothing more than a giant, delete-proof, global filing cabinet for human opinions. Raw data sits there forever, untouched. Every single tool, app, or view that people actually look at gets to filter it however they want (your GAC method, block lists, reputation, whatever). The filing cabinet itself never gets cleaned or sorted by us.
We think the first killer use is replacing every crooked review site on earth (restaurants, doctors, politicians, you name it) with one place that can’t bury or delete what you say.
Would love to hear any ideas from this crowd on solving the chicken-and-egg adoption problem without ever touching the raw data.
Current site (still rough): https://kaosnow.com
Thanks for the work you all do,
Brian Charlebois
Really interesting work, thanks for pushing forward with these ideas. Have you also considered that communities might want to "reverse" some of the learning of an AI? For example, a community is really happy with their AI but collectively feels some small changes are needed, but having made them the model is now worse. Is there a need to be able to revert to a previous state? This might allow for de-risking prototyping new input and exploring alternatives.