Lune AI (YC F24)

LLMs hallucinate on material outside their training data, which often leads to inaccurate answers to technical questions related to code. Whether it’s working with a new library or the ever-changing nature of software documentation, every developer has wasted time on hallucinated LLM outputs. To address this problem, Lune AI (YC F24) is a replacement to standalone LLMs for coding tasks that minimizes hallucinations with a marketplace of expert LLMs trained on hundreds of technical topics. It outperforms the accuracy of standalone LLMs by 37%. You can find a marketplace of community-trained Lunes on hundreds of the most popular libraries and Github repositories for developers. Users can also receive monthly payouts based on their contributions to Lune outputs and the popularity of their public Lunes. Through Lune's API, you can access Tycho, their proprietary automatic context-switching model, and all public Lunes. Users can integrate with any IDE, tool, or client that supports OpenAI compatible APIs like Cursor and other AI client SDKs with the API. Project keys with the API can be used to easily integrate backend applications and workflows where superior accuracy and minimized hallucinations for technical topics are necessary. Founders Josh Park and Sam Park see Lune AI becoming the standard for how developers utilize LLMs for code and technical use cases. Congrats to the team on the launch! ?? https://lnkd.in/gxKw2YGN

It’s been two weeks since I announced that my brother, Josh Park, and I are building Lune AI (YC F24), the layer between a stock LLM and developers, to get more reliable technical-outputs for code and more.?We’ve been closely listening to your feedback so we’re super excited to share some huge announcements/product updates. ? We now support integrations with other AI copilots and tools like Cursor, Continue, and more. You can now use Lunes/Tycho directly in your IDE as a replacement to stock GPT-4o or Claude 3.5 Sonnet, and with a click of a button, continue your chats on Lune Web. ? We’ve made huge improvements to Lune Web to make chatting with multiple Lunes at once more ergonomic. With one monthly subscription to Pro, you don’t have to worry about your own keys to Claude or GPT. ? In addition to web sources, we now support public Github repositories as a Lune training source. Private Lunes trained on private repositories are coming soon for Pro users. ? Our API is now live for all Pro users. You can now call Lunes and Tycho programmatically with our OpenAI-hot-swappable chat/completions API. ? All Pro users now have access to our new payout programs. Two ways you can earn payouts. First by creating public Lunes where earnings are based on the Lune’s popularity. Secondly you can get paid by contributing to Lune’s answers directly in chat to help improve its accuracy. Amazing to see early users in our trial program already accumulating payouts. All payouts will be redeemable starting December. We really look forward to seeing how all of you interact with our platform! Feel free to leave feedback or questions down below, or reach out to me directly.?

Hey everyone, stoked to announce I’m working on Lune AI (YC F24) with my brother Josh Park. Stop using ChatGPT or Claude. Hallucination of LLMs is probably a problem all of you have experienced. Try asking ChatGPT any specific, technical question and there’s a pretty good chance it’ll make something up. If you’re someone who writes a lot of code and utilizes ChatGPT or Claude, this is especially a large problem for you. How many times has ChatGPT told you to implement a deprecated function or made up a package out of thin air? This problem also isn’t just constrained to coding… So we’re building a platform where you can take a URL to any knowledge source (like Langchain’s documentation or your CS class’s online textbook…) and train a “Lune”, to get accurate, reliable, up-to-date outputs with direct references to sources. Early users have seen significant accuracy improvements compared to GPT-4o or Claude 3.5 Sonnet when using Lune for libraries like Numpy or Pandas. And recently, we launched our Tycho-preview model which is capable of automatically context-switching between individual Lunes, so you can talk to multiple Lunes at once. So start using Lune for free today, and check out the many Lunes our early users have made or go create one yourself! We’ll also be rolling out Lune powered by GPT-o1-preview in the upcoming days, only available to premium users. Oh yeah, and Lune premium is half the cost of ChatGPT Plus. We’ve also had a lot of people reach out and ask about Cursor integration and PDF / file support. If this is something that interests you, comment or message me and if enough people are interested we will make it happen soon.