Of course, it’s the community’s call on when and which solution. To help with the assessment, some of the question we want to ask is as follows:
What’s the accuracy. Easily testable with some humans in the loop?
Does it work with my core channels such as Discord, Website, and email?
Can it hand over to a human agent when it can’t answer?
What kind of actionable analytics does the product provide to improve the response quality over time?
How easy is it to maintain the system and retrain the bot with new documents.?
How secure is it?
You mention implementations for other DAOs. Can you provide specific examples, by DAO, of these implementations? Comparitively, what were the prices charged for those engagements?
Is a chatbot a feature or a bug? Anytime I’m greeted with a chatbot, I ignore it as my brain is trained to do.
Further, the scope and scale of ApeCoin DAO, and it’s active community of mods and working groups, doesnt jump out as an obvious fit to me relative to bankless DAO or new chains where many, many farmers and bots love to roam.
The information it obtains is not private. It doesn’t obtain any information rather answers the question that user asks.
In the process of answering the question, if the community wants it to fetch onchain statuses, it can ask for the wallet address and transaction hash to get their status.
Your topic will be moving to the AIP Draft phase in less than 24 hours. Are you content with the feedback received or do you wish to extend community discussion for another 7 days?
If we do not hear from you within 48 hours after your topic closes, your topic will be moved straight to the AIP Draft process.
Thank you for your ideas [and the ApeCoin DAO community for the thoughtful discussions]. A moderator will get in touch with the author to draft the AIP in the appropriate template. Once the AIP is drafted and meets all the DAO-approved guidelines, the proposal will be posted on Snapshot for live official voting at: Snapshot
Follow this Topic as further updates will be posted here in the comments. @helixashu please see your messages for the next steps.
@helixashu , has responded to our questions and has provided consent to share them in this forum for the community.
Yes, the foundation is expected to work with us to make it successful. Here are things we would need from the foundation at each Milestone:
Milestone2: Help us identify the locations from where the bot could learn from. It could be forum, any specific notion pages, Discord conversations etc.
Milestone3: A team of 2-3 persons or more tests the bot and gives the early feedback to fine tune it further. At this point, the bot’s response quality improves by over 50% from earlier.
Milestone4: Deploy the bot in ApeCoin’s discord server. No coding needed.
No
No Web3 connectivity is needed for the bot to function. As such, we don’t have a security audit. From an overall governance perspective, we are going through a SOC2 compliance process which we are expected to complete by the end of the year.
This AIP is aimed at helping the DAO members get accurate, helpful responses quickly. As such, we will track following metrics for the success:
#Engagement with the solution: Measured by #conversations with the bot, #AIPs submitted. Both should be higher than the current baseline. I would love thoughts from the team on how else could we track for this community
Percentage helpful responses: should be 60% or higher.
Average response time: Should come down to less than a minute
We have already provided a sample bot which is Milestone 1. At this point, we take the rest. However, if the community wants, we can split it into two tranches, first half before beginning Milestone 2 and the final before Milestone 4.
PS: After listening to the community’s feedback, we switched to usage based model and reduced our ask from 9000 APE to 2500 APE.
Just view permissions for everything else except write permissions in specific Discord channels where we are deployed.
We collect the following information: Discord User Id (for users who ask questions), Question Text, Answer Text, Thumbs/Thumbs down activity on Helix responses.
Sure thing.
A DAR package is being worked on and upon completion this AIP will move into Administrative Review. Follow this Topic as further updates will be posted here in the comments.
As a bot, it is expected to have a response time in seconds. I would consider this as the baseline, not a “success factor”.
This is a limiting factor on the LLM itself, not on bad architecture. See here for response times across the major LLMs. When OpenAI shows the … or some chat bot says it’s “thinking”, it actually is just processing it. We don’t have good LLMs with sub second response times which we can shuttle to the user in seconds.
I am aware about LLMs and I have also developed a discord bot leveraging OpenAI.
By seconds, I did not mean 2-3 seconds. I meant that talking about a response time of “less than a minute” as a success factor feels like stat padding. Barely anyone serious makes these AI chatbots with a response time in minutes.
@helixashu has responded to our questions and has provided consent to share them in this forum for the community.
A DAR package is being worked on and upon completion this AIP will move into Administrative Review. Follow this Topic as further updates will be posted here in the comments.
This Topic has been rejected based on the DAO-approved guidelines due to no response in the last 30 days. The Topic may be submitted again by any user and upon approval, will be open for 7 days for community discussions.
This Topic will move and remain in the Withdrawn AIPs category.