In the world of automation and AI, DAOs have the ability to become one of the most important forms of organization. DAOs use proposals among decentralized groups to make crucial managerial decisions, and the use of LLMs (Large Language Models) could be a game-changer for them.

‍

In this article, we will take a brief look at DAOs and dive into how LLMs can help them. If you are already familiar with DAOs, feel free to skip ahead here.

‍

First off, what are DAOs?

‍

A Decentralized Autonomous Organization - or a DAO, is an entity structure in which token holders participate in the management and decision-making of an entity.

‍

The keyword when it comes to DAOs is “decentralized”. Here decentralization essentially means that there is no specific central authority that holds controlling power; instead, the power of authority is distributed.

‍

The power of DAOs lies in the fact that the trust is placed in the code itself rather than in the individuals and their reasoning to make decisions.

‍

The code is publicly available and fully transparent, allowing anyone within the organization to contribute ideas and introduce new narratives that may not typically be seen in a centralized organization.

‍

How do DAOs work?

‍

In a DAO, decisions are made from the bottom up, often through the ownership of a blockchain token.

‍

DAOs operate using smart contracts, which are lines of code that automatically execute whenever a specific set of criteria is met, and these smart contracts establish the rules for the DAO.

‍

Those who have stakes in a DAO will have voting rights and will therefore be able to influence the organization’s direction or decision-making. However, a proposal can only actually pass if the majority of stakeholders approve of it.

‍

Essentially, DAOs are built through three steps:

• Creating the smart contract
• Funding the DAO
• Deploying the DAO onto the blockchain to be used for decision-making

Once a DAO is deployed, proposals are needed to initiate governance, and this is where Large Language Models could make a significant difference.

‍

Generating Proposals for DAOs using LLM: An Experiment

‍

In order to encourage stakeholders to start voting on your ideas and suggestions, your DAO proposals need to be very clear, detailed, and well-written, providing as much information as possible.

‍

For a fully transparent experiment, our expert AI/ML team at Linum Labs conducted an experiment to generate proposals for DAOs using LLMs.

‍

The story:

‍

Using data from Snapshot and existing DAOs, our experiment focused on creating proposals for individuals establishing their own DAOs. Based on the Dao's purpose and function, we were able to generate proposals for them.

‍

The goal for this was to generate better DAO analysis, representation, and grouping.

‍

How did we do it?

‍

We used the existing data in our database, specifically the relevant descriptions and top proposals, and fed them into an LLM tool to generate summaries, ideas, and proposals.

‍

The main focus here was to steer the LLM toward specific domain knowledge distribution to obtain the most relevant results.

‍

Prompting and generating

‍

In the prompting stage, we asked the tool to generate the proposal title, body, and choices of a promising proposal aligned with the given information such as space information (describing the organization's goal), space categories (the organization's type), and proposal ideas (considerations of the proposal creator).

‍

The purpose of the proposal is to present an idea within the organization and allow the participant to vote.

‍

We used some examples from Aavegotchi, and provided our input, including space information and proposal ideas, to OpenAI.

‍

We, then, compared the original output for this proposal with the LLM-generated one and found the LLM-generated one to be more specific and concise.

‍

Our criteria for selecting proposals include proposal success, DAO popularity, and similarity to user DAO specialization. We are training an NLP model to process this information and return the samples, which we then summarize into key points to embed this knowledge in LLMs.

‍

Here are some examples of the sample:

We also experimented with different prompts, such as summarization and idea generation.

‍

Summarization

‍

For the summarization prompt, we provided the tool with a very straightforward, role-defined case and asked it to summarize the proposal to be short, informative, and even parsed with JSON.parse().

‍

Idea Generation

‍

For the idea generation prompt, we also gave the tool a very straightforward case defining its role, experience, the DAO it works for, and what it specializes in. We requested it to generate proposal ideas based on similar DAOs and include motivations for each proposal idea.

‍

The prompts we used for this stage were as follows:

The output was a specific, data-driven generated proposal with references and motivations.

‍

Here’s an example:

The result?

‍

Lastly, We ask the LLM to generate a complete proposal structure based on the provided idea using the following prompt:

And below are samples of the generated proposals:

By using the LLM to cluster the data, we were able to generate around 20 topics from the 14.8K DAOs we used in the experiment.

‍

These are examples of the topics we clustered:

In addition to the many functions of GenAI tools, using them to generate proposals for your DAO will save you a lot of time and effort, increasing the chances of your proposal being seen and approved.

‍

There is still a long way to go for AI and DAOs to evolve individually and together, but we hope to see further innovation in this intersection. We would be happy to discuss this project, AI, or Web3 further with anyone who might be interested. Feel free to reach out to us directly at linumlabs.com.