[READY TO VOTE] AI Assistant for governance support Mission Request

Governance Fund Missions 🏹
, ,
1

AI Assistant for governance support Mission Request

Delegate Mission Request Summary:
This mission requests the construction of an AI-powered assistant to generate guidance and digested information about the Optimism Collective for interested individuals.

S5 Intent: 4 Governance accessibility

Proposing Delegate: SEED Latam delegation team

Proposal Tier: Ember Tier

Baseline grant amount: 8,000 OP

Should this Foundation Mission be fulfilled by one or multiple applicants: Up to 2

Submit by: To be set by Grants Council

Selection by: To be set by Grants Council

Start date: ASAP

Completion date: Likely to be in 4 months

Specification

How will this Delegate Mission Request help accomplish the above Intent?

Keeping up-to-date with the complexity, speed, and variety of topics covered in governance can be a daunting task, causing both current and potential interested persons and entities to delay their full involvement while they try to grasp the context. To attract more interested parties to participate in the discussions and opportunities that governance provides for the future of Optimism, and to combat voter apathy, AI tools can be used to simplify the interested party’s experience and speed up their response to extensive information.

So, in order to facilitate knowledge sharing and discussion on Optimism Collective’s forum, we propose building an experimental product that provides guidance, digested information, and updates on the governance & forum’s structure, processes, and latest news. This product will be aimed at individuals interested in governance and two key features should be built: a chatbot and a summarizer.

Brief description:

  1. Chatbot: such as GPTs, this AI-powered assistant uses information from the forum and key documentation on Github to answer questions in a chat format. It should be updated regularly to stay current with governance.
  2. Summarizer: A feed of AI-generated summaries for key governance topics, including updates, announcements, proposals and more.

What is required to execute this Delegate Mission Request?

We are looking to invite teams who are enthusiastic and have the following background:

  • Experience and knowledge of Optimism governance, ability to be extremely rigorous with quality control
  • Deep understanding of GPTs and how to build them
  • Experience building a clean user interface

Additionally, the following deliverables should be taken into account:

  • General design: This includes the design of products, data flow, and UX/UI with user flows. This is conceived as an external product.
  • Development of the AI engine: Whether based on OpenAI or any other platform, it should be adapted to the data sources available for Optimism governance and response logic. The AI engine and its prompts should be tested accordingly. We encourage considering open source models in this regard.
  • Forum integration: The product should be able to extract relevant information from the original sources such as the forum. This could need the API from external sources or alternative methods. If it is not possible to access the API during mission execution, the proposer should be able to track information generation manually and stay updated on AI queries.
  • Comprehensive documentation: The logic behind how the whole product works should be open to the public. It should come with a corresponding FAQ and technical details that enable others to contribute.

How should the Token House measure progress towards this Mission?

  1. Engine Development: This involves integrating AI technology with the relevant data sources and training it to perform effectively.
  2. Deploying a PoC: the chatbot is publicly available and ready to provide users with prompt and accurate responses.
  3. Final Frontend: Created the external instance to host information from Summarizer and Chatbox.
  4. Content generated: number of summaries generated by the Summarizer.
  5. Response: Evaluation of the Chatbox could be conducted by testers (delegates by example) to ensure proper basic accuracy and coverage, at the end of the period.

How should badgeholders measure impact upon completion of this Mission?

  • Staying updated: The product has been updated with the latest news and activities on the Optimism Collective.
  • Content/response quality: Evidence of the product provided precise and adequate information at all times, on any topic.
  • Usability: By number of users, and retention.
  • User stories: Where the success of onboarding a new contributor to the collective was highly dependent on the product.

Have you engaged a Grant-as-a-service provider for this Mission Request?

No.

Has anyone other than the Proposing Delegate contributed to this Mission Request?

SEED Latam delegation proposed the mission request. Special kudos to our contributor @Jadmat for working with me on this.

New: After discussing a similar Mission Request proposed by @Michael, we collaborated and incorporated relevant ideas and feedback into this proposal.

Background

In the past, various efforts have been proposed and implemented in the field of governance. Some of the most relevant ones include:

  • MakerDAO governance with MIP106: Support Scope Bounded Mutable Alignment Artifact; Governance Artificial Intelligence Tools (GAIT): “The GAIT consists of multiple redundant, diversified large-scale artificial intelligence models operating in remote data centers, trained to provide alignment engineering and business process optimization for MakerDAO, SubDAOs and Ecosystem Actors implemented as Scope Artifact generation and strengthening. They are accessible by Alignment Conservers, MKR holders, and SubDAO token holders”. You can find more information about this product at https://makerchat.org/.

  • Aave governance with x23, an external project: This is an external project that offers a dashboard which simplifies Aave’s governance activity. It provides a summary of each proposal and discussion space. You can find more information about this project at Aave Discussions.

14 Likes
2

This is a very first draft that seeks to implement AI technologies for the benefit of OP governance.

As a draft, feel free to provide feedback from any perspective; we are open to receiving suggestions for changes and including collaborators more involved in this field. So, while the scope and vision look great, many changes can be made to either aspect.

3 Likes
3

Made some comments on Discord about the proposal: Discord

TL;DR:

  1. I believe this proposal is a great idea and will greatly benefit the optimism collective but it needs to be very clear what’s in scope and what’s not.
  2. I think the proposal should be split into sub-proposals. IMHO building a chatbot is quite different from building a Meetingbot and asking one team to do all of that is too much given the baseline grant amount.

Cheers,
Maxime

8 Likes
4

Sounds like a interesting, potentially useful idea.
One of the most important thing is that this tool would actually be used by the broader Optimism community. I’d be a shame, and sub-optimal, that a tool like this were to be approved but it was not used.
Additionally, I agree with @MaximeServais comments regarding splitting them into 2 proposals: The chatbot and the workerbot

2 Likes
5

Hi @Joxes and SEED LATAM delegation team,

Utilizing the AI technologies to further improve the governance process and supporting delegates (especially for newly joined members to catch up with what’s happening within the DAO) with the outputs by them is definitely helpful for the Optimism Collective.

As you mentioned one for Aave, x23.ai now has covered Optimism too The Optimism community | x23.ai. There are great products already for the general purposes like summarizing zoom calls (e.g. Otter.ai) We believe the Request particularly focuses on initiatives that needs detailed customizations and gives great value. Based on that, we would rather strip areas that can be covered by the external products, but focus on the initiatives described above, than splitting them into 2 proposals.

4 Likes
6

Hey all!

Thank you all for your valuable feedback. We have made some changes to the mission request to better align it with our goals:

  • We have removed the workerbot and meetingbot as they were deemed to be of low priority. Our focus now is on a unified product that includes the chatbot and summarizer, as both are important components that should not be separated.

  • We have provided better measurement guidelines for the consideration of Token House and badgeholders.

Regarding the baseline amount, we are still open for more feedback. Ultimately, we believe that the applying teams should have their own measures to ensure that costs do not become unprofitable. This could include limiting the number of queries or update times. We want to encourage creativity among the applying teams as they work to find the best strategy for optimal use.

@Tane @dmars300 @MaximeServais

4 Likes
7

I would like to see this pushed to the next stage, so I am an Optimism Delegate with sufficient voting power and I believe this proposal is ready to move to a vote.

However, there is a part I find concerning:

This is just my personal opinion, but personally, it’s important to make sure that we build products that are resilient, and thus ideally do not require reliance on an API from a centralised web2 company, especially one that can ban API requests (they could blacklist web3-related requests) and deprecate models.

The open source, self-hosted AI space has developed a lot in the past few months, especially from releases by Mistral like Mixtral. Even if using the models on a centralised server, if the appropriate prompts and even finetunes are shared anyone could easily run their own version at home.

Best of luck, and really excited to see this!

3 Likes
8

Is it possible to accomplish this without requesting Foundation action or support like providing a forum API key?

I don’t want to be the bad guy but if not this could be violating a rule. I want this to happen just asking about the implementation process.

2 Likes
9

Hi all, I’m the founder of x23.ai, which was mentioned in this thread.

I’m a big believer in experimentation and think more people/devs should keep building and deploying AI enabled tools. However from my experience building and bootstrapping x23, a light weight MVP can be created quite quickly and effectively with a much smaller budget.

My suggestion would be to build specific AI tools for the Optimism community, such as MakerDAO is doing with their GAIT tools that are specific to the MakerDAO community. Maybe OP have certain long term goals around governance participation, grant distributions, or automated proposal feedback?

Some of the more general use cases such as chatbots, discord plugins, summarisers etc are already in development by us and others, so it could be a waste of DAO funds IMO.

For example, this is the summary of this thread: [DRAFT] AI Assistant for governance support Mission Request | Optimism. You can also search for all other OP discussions here: Optimism Discussions, as well the snapshot vote proposals and OP code repositories (see navigation in top right).

6 Likes
10

Big fan of an AI tool. I think it is worth building a conversational agent to ask questions about governance (I had proposed a similar one)

That said, I agree with the feedback on this proposal and on mine that 30k OP is quite large given the large number of tools. I would be happy to support this proposal if that amount is reduced! I ended up coming down to 8k on mine, which I think is reasonable.

Also, I think its important that the trained model is available for others to adapt from. I think using a GPT should be allowed, it the quickest path to a working product, but there are definitely bonus points for getting an open source model up and running.

Edit: The BIGGEST challenge I see with this project isn’t getting a chatbot working, its getting one that accurately responds to governance questions. It should be able to answer questions like:

  • “What is the process for submitting a mission request?”
  • “What are the important dates for delegates for Season 5?”
  • “What are the S5 intents, and why are they important?”

I strongly encourage anyone who is considering applying to this mission in the future to have a strong plan in place for quality control.

3 Likes
11

Hey! Glad to see you here.

I agree that the GAIT tools developed by MakerDAO are worth experimenting with. However, we should proceed step-by-step and not rush into implementing tools like this in governance processes unless there is a pressing need, here I personally I would take a more conservative approach in this season. As a first step, I believe this mission request serves as a good test of the team’s ability to design a responsive and attractive tool, while also seeing how it is received by the community members.

When it comes to other solutions, I think it’s best to consider it as a product and avoid fragmenting its features to ensure maximum attention. Therefore, we should keep summarisers, chatbots, and any other complementary AI features in a single package.

2 Likes
12

I believe the System you are describing should benefit 1 just as much for all , this should be, thought of , we should skip to the moonshot,train ai, test and contain it properly, errors and mistakes with contracts tokens and chain possibly even blockchain data validity … all this if legally viable and possible

1 Like
13

I strongly support this Mission Request and agree on the benefits of keeping it as a product instead of fragmenting features into smaller products.

1 Like
14

This could be interesting and I would like to see specific proposals for this mission so:

I am an Optimism delegate with sufficient voting power and I believe this proposal is ready to move to a vote.

15

I am an Optimism delegate with sufficient voting power and I believe this proposal is ready to move to a vote.

16

Thanks @Joxes and others for improving the Request for review.

We still believe that applicants can propose products that will make good impacts according to what the Intent 4 aims to achieve and we will carefully look for Mission Applications based on this Request if it’s passed.

We are an Optimism delegate with sufficient voting power and believe this Request is ready to move to a vote.

1 Like
17

The Grants Council has opened early submissions as an Indication of Interest for this mission request here

For your application to be considered, the Mission request must pass the Token House vote on February 14th. Early submissions will not be considered if a Mission Request is not approved on the 14th.

1 Like
18

Like the proposal a lot, and I think the reduced amount is reasonable.

I am an Optimism delegate with sufficient voting power and I believe this proposal is ready to move to a vote.

19

I can’t emphasize enough how important is the data availability for an AI project. I would recommend a small scale mission request for an API to be started now.

Also model provider SaaS companies (OpenAI, Anthropic, etc.) have reason to downsize their models. This tends to create behavior and performance changes.
To decrease the TCO(total cost of owneship) periodic development activity should be avoided. I request you to augment your proposal so it requires an open source LLM to be used as the AI engine. Competitive open source alternatives are available already!

2 Likes
20

I like this idea and concept, but I have a couple of questions/concerns before I give my onchain approval.

Was there ever any clarity regarding @Gonna.eth 's question on whether this requires Foundation approval for api access to the forum? ccing @maxwell @lavande

I think the most important aspect of this project will be the ability for others to build on top of it. I’m not confident that whoever completes this mission will create a product that’s put into great use by the collective, but I do think that the underlying infrastructure built will be useful to a project that is used widely by the collective. Whoever takes on this mission should have a strong plan for documentation and making the project easily iterable by anyone.

1 Like