Ahoy ICP folks,
The Foundation voted to REJECT motion #133401, and as someone close to the matter, I will explain why we made this decision. I will try to be as transparent as possible while giving not overwhelming context.
Background
The motion proposed that InternetComputer.org replace a SaaS provider for #AskAI (a RAG service) that runs on Web2 with the Kinic Vector Database (an ICP provider).
Where We Agree
The proposal correctly states:
“At the time AskAI was installed (October 2023), achieving it on-chain was difficult, so we understand why it was necessary to use a Web2 company.”
We agree with the following points:
-
At that time, we faced limitations.
-
A full Web3 solution would be preferable.
-
Vector databases are a feasible solution within the ICP ecosystem.
Where We Disagree
Despite considering the proposal reasonable, we found that our disagreements outweighed our agreements. It comes down to three reasons:
-
Kinic does not meet the needs of the website.
-
There is a larger opportunity for Kinic and the ICP ecosystem.
-
Motion proposals for changing providers could lead to unhealthy relationships between proposals and providers.
1. Kinic Does Not Meet the Website’s Needs
Kinic does not meet our current needs (Fwiw,you may address me personally, as I was the main decision-maker on this matter). A vector database is only a small part of what DFINITY requires. We need several other essential tools surrounding the database. If we switched to Kinic, we would need to develop or create many additional components, such as:
-
Regular cron jobs to check sources and update the vector embeddings in the database, ensuring information is up to date (as done by Kapa.ai).
-
Admin dashboards to allow DFINITY’s R&D and customer support to review questions, improve AI training, and manage user support.
-
Metrics tracking the types of questions to help us improve documentation.
Ultimately, building and maintaining all these features would be a drag on the main goal of DFINITY: R&D.
2. There Is a Larger Opportunity for Kinic and the ICP Ecosystem
We believe that a better opportunity for Kinic and the ICP ecosystem lies in Kapa (a SaaS wrapper that uses multiple types of vector databases) incorporating Kinic as one of its vector DBs.
This approach would:
-
Provide Kinic and other ICP ecosystem components with real use cases, driven by economic incentives. Kapa could choose whether to integrate with Kinic based on the economics and its customers’ preferences (e.g., we would prefer Web3 vector database hosting). We have communicated this to Kapa.
-
Help Kinic improve by working with entities outside the ICP ecosystem, particularly those from the Web2 space.
For transparency, I initiated discussions between Kinic and Kapa, but these discussions slowed over the summer. I could have done a better job managing this relationship, and I see this motion proposal as a reflection of that.
3. Motion Proposals for Changing Providers Could Lead to Unhealthy Patterns
Lastly, I advocated for DFINITY to reject the proposal because motion providers asking ICP entities to switch providers could set an unhealthy precedent. To clarify, consider a hypothetical example where a motion proposal suggested, “Kinic should switch to DiegoTech, a Web3-based code editor.” Using motion proposals in this way might deviate from the intent, which should ideally focus on advancing the Internet Computer. While I can see how motion proposals for improving NNS dapps fall under this scope, requesting a provider change for a website does not, in my opinion, meet that standard.
Conclusion
Are we ready to change our mind on this? Absolutely!
Are we ready to answer more questions? yes!
Please do!
Thank you, all.