Hi all. I’ve heard some questions recently, and second hand from the watercooler chats and discussion, so I thought it would be good to clarify the Livepeer AI SPE Milestone and what we hope to accomplish.
As a reminder, and as described in the treasury proposal, the Livepeer AI SPE was created to bring AI Video compute capabilities to the Livepeer network. The initial proposal requested funds for the first milestone of technically validating that this was possible on the Livepeer network and proving the network’s cost effectiveness and reliability at performing these tasks.
What is the most important outcome of this milestone?
It is being able to show the world evidence that the Livepeer network now can perform generative AI jobs.
- Orchestrators can perform these jobs and receive payments
- Potential users of these jobs have an open access point to use the network to perform them
- Ideally it is highly cost effective
- Ideally it is highly reliable and performant (though we of course will need to iterate on this over time)
What is the strategy for doing this?
In very summarized form:
- Quickly build these new job types into a forked version of the Livepeer node
- Launch them on a subnet or O’s and Gateway nodes that are running these forks, rather than wait to merge them into the main production client.
- Work closely with a design partner application (reasons emphasized below in next section) that is subsidized via the treasury
- Collect the data and evidence about the cost and performance of the subnet with enough volume, and package it up into a case study to tell the world about Livepeer’s new capabilities.
The above is the fastest path to completing the milestone. This sets us up for the next step, which is to use all of the learnings from the above process, and newfound readiness to support more users, to define the path for productionizing this, extending it’s capabilities, and considering possible go-to-markets and ecosystem growth efforts.
Why fund a consumer app?
First, it’s worth pointing out that launching a hit consumer generative AI video app is not the big story here. It’s not the goal of the SPE. It’s not nearly as important as the Livepeer network being a scalable, open, and cost effective infrastructure for AI video. But there are a few very good reasons to have one being built, and have its usage be subsidized as part of the first milestone. In order of most important reason to least:
-
It’s a high touch design partner, and that dramatically accelerates the development of the AI capablities on the subnet. Two accomplished builders working full time to actually use test, iterate, and use the capabilities have revealed countless requirements, discovered countless bugs, and informed countless “middleware” needs that have dramatically sped up the process of getting the AI subnet useable.
-
It creates something visible to create tangible outputs for the market to absorb when we show off Livepeer’s new capabilities to the world. Not only is the app itself demonstratable, but the outputs are sharable and interesting. and can be used in the marketing of the new capabilities.
-
It enables data collection. By making a consumer app free to use, as subsidized by the treasury, there’s actually a fast chance of getting enough usage to enable price competition and scaled usage of the subnet. This allows us to collect data and market the case study about cost and reliability in a short period of time. If we had to execute a B2B motion and get 3rd party app integrations it could take a very long time to get those integrations into production and actually produce evidence. We’ve seen that if consumers have the opportunity to type prompts for free and experiment with outputs, they will.
-
There’s a chance it could go viral! While there’s still a lot of work to do with the app polish, UX, viral mechanics, etc…that’s an iterative process. There’s no expectation that it is the next Giphy, but there’s a small chance it could be, and it’s great to see two entrepreneurs who can run with the iteration towards that if there’s positive signal. For a typical entrepreneur, building your app on a non-existent underlying tech platform, that is buggy, sparse in features, and on an unclear delivery timeline is a huge risk. I’m grateful that there’s a team that believes enough in Livepeer and LPT (at a time when the value of said LPT was lower) to spend 4 months working in on an experiment so risky.
But what about other apps? Should they be funded or subsidized?
It is awesome to see other people building direct on the early AI capabilities already. We want a ton of that! So yes, I definitely think that if they’re committed to iterating, supporting users, and have a plan to try to grow and generate usage, they can and should be considered to access some funding through the SPE (or subsequent SPE proposals).
The initial SPE milestone didn’t create a subsidized builder’s program. It focused on ensuring one solid design partner to accelerate the build process for the network, and ensure data collection, and something marketable to show off as described above. But a future SPE milestone certainly can create this funding and program. Which brings me to the last point for today 
There’s a lot more to come!
We’re still diligently working to complete the first milestone, launch an open subnet, launch the demo app, collect the data, market the results, etc. But as we are getting closer to achieving this, I wanted to remind everyone there’s a lot more work to do and more milestones ahead! Look for the SPE (and potentially a modified version of it with additional familiar faces) to propose subsequent milestones related to productionizing the capabilities onto our main network, adding support for more job types and workflows, and creating ecosystem growth programs such as builder and credits subsidies, grants, tools, and more!
It’s great to see how much of a community effort this is, and via the growth of the treasury and grants programs, I think we have a lot of opportunity to put those LPT to work to really grow out a flourishing AI video ecosystem. Let’s do it.