Delegate Incentive Program Removal Wave #1

Introduction

Following the provisions of Arbitrum’s DIP proposal, this post aims to announce the expulsion of the following delegates: @kuiclub and @duokongcrypto (previously Kuiqian). This decision is based on behaviors that violate the rules and principles of the program, including suspicious activities, inappropriate conduct, and participation simulation.

According to the “DIP Ban” section of the DIP proposal:

  • The program administrator will have the right to expel a delegate if they attempt to game or exploit the program or if the delegate does not meet any of the aforementioned requirements to be considered eligible. This decision is at the discretion of the program administrator. In all cases the ban is permanent.
  • The affected delegate may request a Snapshot vote to ratify, change (for suspension), or revoke the Administrator’s decision. This serves as a one-time appeal, and the decision made by the DAO will be final.

Based on this framework, it is argued that the mentioned delegates fail to meet the eligibility requirements and/or have engaged in activities that justify their permanent expulsion.

Behavioural Patterns

  • Suspicious Activity: Both delegates commented on the same proposals within a very short time frame of each other, and their posts contain similar content.
  • Suspicious External Connections: There is evidence of coordination between these accounts, indicating a possible conflict of interest or manipulation of the program.
  • Bot-like Behavior: This delegate’s interactions in the forum reflect automated behavior, with generic responses and a lack of qualitative analysis. There is clear abuse of tools like ChatGPT for generating comments, instead of using it solely for translations or grammatical improvements. They also recently commented on old threads without a logical reason to do that.
  • Low Value Comments: both delegates usually post low-value comments as can be seen in the rubrics of the last few months.

Previous Actions

Back in August, during the first interaction of the Delegate Incentive Program, we initiated the first conversations with Duokongcrypto regarding the excessive use of artificial intelligence tools to generate comments.

In November, after noticing the mirrored behaviors previously mentioned, we sent both delegates a message through the forum and created a joint Telegram group. These measures were taken to establish a more fluid communication channel, allowing us to discuss how they could improve their contributions to the DAO and continue progressing toward the professionalization of their work.

We expressed our concerns to both delegates, requested explanations, and suggested improvements.
After these conversations, we noticed a slight improvement in both delegates’ contributions, and their “mirrored” behaviors ceased. We continued monitoring their activities and even responded to several members who approached us with similar concerns.

However, during January, we observed an evident decline in the quality of their contributions, primarily due to the excessive use of artificial intelligence tools.

Administration Decision

Based on the repeated patterns of suspicious behavior, AI-generated low-value contributions, and mirrored activity between KuiClub and Duokongcrypto, we strongly believe that their participation in the Delegates Incentive Program (DIP) of Arbitrum DAO should be terminated.

Both delegates were previously given multiple opportunities to adjust their approach and improve their contributions. We actively engaged in discussions with them, provided direct feedback, and even created a dedicated communication channel to help them refine their participation.

Despite these efforts, their behavior not only failed to improve in a meaningful way but has regressed over time. The resurgence of AI-generated comments, automated activity, and mirrored interactions indicates a continued disregard for the quality standards and principles of the program. Furthermore, this pattern raises concerns about the genuine nature of their engagement within the DAO.

Given the evidence presented, KuiClub and Duokong will be removed from the DIP, as their continued participation undermines the program’s credibility and effectiveness. While we believe in fostering inclusivity and growth within the governance framework, it is essential to ensure that delegates engage meaningfully, thoughtfully, and professionally to uphold the integrity of Arbitrum’s governance.

This measure is necessary to preserve the program’s legitimacy, protect the community’s interests, and maintain the high standards expected from participating delegates. However, this does not mean that both will cease to be Delegates, they are free to continue participating in Arbitrum DAO but will not be eligible to receive incentives from this program.

18 Likes

Hello!
Relative new to the forum as a commenter, but I have been reading it for a while.
It’s clear that some delegates only use chatgpt and similar tools to draft comments, whoever uses them can recognize the pattern. Maybe is worth asking, what is a rightful use of these tools? I know that some use it for translations for example, which should be a legit use case. But is quite difficult to distinguish between a translation, and just pasting a thread with a prompt to create an answer, which should be punished.
One related question is

how is the “excessive use” defined in this case?

In general, I support the organizers and their decisions.

They’ve clarified on it in the section below

1 Like

Let me add 2 cents here. I obviously might be wrong in this, i see plenty of chatgpt comments in the forum. It’s a bit difficult sometimes to distinguish between:

  • comments generated by simply throwing a whole proposal in claude et similar with the prompt “comment on it”
  • comments generated by throwing the whole proposal and give a prompt like “give an answer that agrees to the proposal”
  • comments generated by throwing the whole proposal and give a prompt like “i think X, Y and Z about this proposal, build a comment around”
  • comments generated by writing in a non english language a list of bullet points, and translated through AI in english
  • others.

So I think raising the question about what is the legit usage is a good thing.

I honestly don’t like AI comments in the forum. I also don’t like people who don’t do the effort to overcome the english barrier (this is coming from a non native english person as you might understand from my broken writing from time to time). I can also be forward looking enough to understand that AI communications, AI interactions, AI agents are indeed the future, and trying to gate the usage of the tools here is like saying in the second half of 1800 that horses are better than cars.

The problem that we saw above, a delegate almost only using AI to interact with the forum adding low value, is something that we will see more and more. And we will see it in a way that will be less and less easy to detect because not only the LM will get better, but also more users and delegates will start using it.

Going back to the roots might help. DIP 1.5 has been about rewarding quality of comments.
Can quality be achieved through AI? Yes.
Can quality be achieved with AI only? Likely no.
In the above we are saying that interactions only through AI, with no proper value add, should be punished with a DIP ban. Does this implies that low quality comments in general, with or without AI, should be “punished” as well? Maybe not, but also not sure.

Maybe this discussion, in the end, could gravitate toward the fact that low quality comments should be rewarded way less than what they are now. With or without AI.

8 Likes

You mixed up the discussion thread and wrote the same point there.

2 Likes

I’ve deleted it. thank you for the reminder

We appreciate the transparency of the @SEEDGov team and the continuous effort to maintain high participation standards for Arbitrum DIP. The timeline of events (starting August 2024) shows that this is not a hurried decision and has been given reasonable time to develop. We also appreciate the clarification on the use of AI by delegates. Evidently, a problem only arises when the use of AI is excessive or leads to low quality contributions. Overall, we support punishing exploitative, low-quality contributions or behavior whether or not AI is involved.

4 Likes