• Subscribe
    1. Discussions
    2. How do you build a moat around GPT3, 4 or any other open...

    How do you build a moat around GPT3, 4 or any other open source LLM?

    Aakarsh Yadav
    6 replies
    A startup or a product usually is more fundable (or adoptable) if there's a solid moat around the product. With the release of more and more open-source LLMs in the future it seems that AI based small startups will produce brilliant products but the defensibility that they build will always be questionable since big players may easily pip their product.

    Replies

    Fabian Maume
    Warmup Inbox
    Warmup Inbox
    That is a good question. I see two ways to build a moat around GPT3 apps: - Use GPT3 with specific data. We integrated GPT3 with QApop to write answers to Quora questions. We have some infra to extract data from Quora, which would be hard to replicate. - Have a good UI for editing. From my experience the output on GPT3 is never perfect: you need some proofreading and editing. You can generate a lot of value for the end user by providing a good UI for editing. For example, offer the option to regenerate text for a paragraph you don't like.
    Share
    Aakarsh Yadav
    Hushl
    Hushl
    @fabian_maume true that. I think lex.page kills it here. They have built a very minimalist app, that just provides fab UI for editing your article / blog.
    Share
    Richard Gao
    Evoke
    Evoke
    There is always that risk, but just think of OpenAI's business model. They're mostly in the business of selling their API for people to build those products. It would be a waste of resources if they did that themselves. They may spruce up the UI for their playground a bit, so the low quality AI writing apps will be knocked out But if you have really good UI, or you're making an analysis tool, then you should be fine.
    Share
    Aakarsh Yadav
    Hushl
    Hushl
    @richard_gao2 that's a good point. The uses cases are also so many that it might be hard for OpenAI to themselves identify everything. And then having a good UI is a whole different game.
    Share
    Adithya Narayanan
    Hushl
    Hushl
    Solid question! Responses off the top of my head: 1. Fine-tune your model: Get enough user data to build a proprietary layer of data on top of your foundational layer to produce superior output in your niche. More personalization, more contextualization in your chosen niche. 2. Network effects: Build a social layer on top of your app to allow users to engage with each other, build community etc. 3. Vertical SaaS: Pick a space (real estate for instance) and integrate GPT3 in the workflow in that space, and build integrations around it. Basically build SaaS that includes GPT3 - but does more than just generate content 4. Superior UI: Just significantly better design. Take a design-first approach, provide a LOT of user delight via well thought out UX and UI, and people won't switch to a poorly designed app.
    Share