Boundary (YC W23)

🚀 We made SF TECH HISTORY at yesterday's SF Demo Night with Product Hunt!! 🚀 Nearly 100 tech startups applied for a chance to ENTER THE ARENA 🏟️ and demo live in front of 250+ attendees, making this undoubtedly our most anticipated event ever! 🤩 🧑💻 FEATURED STARTUPS 🧑💻 🔥 Actual AI – Boost team performance (John Kennedy) 🔥 Toolhouse – AI agents that do things (Daniele Bernardi) 🔥 Scoop Analytics – Automated BI reporting (Janet Gehrmann) 🔥 Coval (YC S24) – Smart agent simulations (Brooke Hopkins) 🔥 Clarity – Call insights made simple (Rohan Varma) 🔥 Structured (YC S23) – Automating data discovery (Amrutha Gujjar) 🔥 OpenPipe – Fine-tune with ease (David Corbitt) 🔥 Perhaps – Personalized site experiences (Gonzalo Enei) 🔥 Boundary (YC W23) – Guaranteed LLM outputs (Vaibhav Gupta) 🔥 CommonAR – Spatial networking redefined (Rohan Vaidya) At the end of the night, we held a community vote for the best demos to be immortalized in the Demo Night Hall of Fame! (check out demos.genaicollective.ai) After nearly 500 votes, here are the winners! 🗳️ 🏅 AWARD WINNERS 🏅 🏆 Best Overall: Coval (YC S24) 🤖 Best Technology: Boundary (YC W23) 🎨 Most Creative: CommonAR This event would not have been the incredible success it was without the active participation of everyone in attendance who submitted nearly 1,000 pieces of invaluable feedback (and 15,000+ virtual claps!! 👏) to all the demoists! Thank you to our outstanding sponsors, Graphite and Predibase, for making this event possible! Finally, a huge shoutout to SHACK15 for graciously hosting us in SF's premier venue! 🌁 This is only the beginning of what is quickly becoming a global network of events that elevates the entire tech ecosystem worldwide. We intend for this to become the most impactful event series in tech history. THANK YOU ALL for being a part of it!! 🚀

What does it mean to stream data out of an LLM? You've likely heard of tokenized streaming. But have you heard of semantic streaming? Full blog post in the comments 👇 Check out the images of the graphs below and see if you can spot the difference? When streaming numbers with token based streaming, you may get something like this: [2 [20 [202 [2024, [2024, 2... But what you really want is: [ [2024, [2024, 2025, ... Because the intermediate state doesn't matter until it complete. Numbers only matter SEMANTICALLY, after all the digits before a decimal place are available. We've built this natively into BAML and make the type-system represent this nuance. Why does this matter? UX is going to be an even bigger differentiator. Waiting on a blank page is infinitely worse than waiting with streaming text. Waiting on streaming text is infinitely worse than streaming semantic chunks. 📣 Greg Hale for inventing this! Boundary (YC W23)

What does it mean to stream data out of an LLM? You've likely heard of tokenized streaming. But have you heard of semantic streaming? Full blog post in the comments 👇 Check out the images of the graphs below and see if you can spot the difference? When streaming numbers with token based streaming, you may get something like this: [2 [20 [202 [2024, [2024, 2... But what you really want is: [ [2024, [2024, 2025, ... Because the intermediate state doesn't matter until it complete. Numbers only matter SEMANTICALLY, after all the digits before a decimal place are available. We've built this natively into BAML and make the type-system represent this nuance. Why does this matter? UX is going to be an even bigger differentiator. Waiting on a blank page is infinitely worse than waiting with streaming text. Waiting on streaming text is infinitely worse than streaming semantic chunks. 📣 Greg Hale for inventing this! Boundary (YC W23)

This Friday, come watch our cofounders talk about how to build LLM pipelines that don't fail! We'll focus on reliability and correctness for about an hour with real code at the end.

If you're interested in seeing what BAML looks like and why it helps, MLOps Community has invited us to come speak. Come by and check it out this upcoming Wednesday.

Its quite funny how polite our issues on Github are... so many start with a thank you and I always feel weird cause I'm thinking "no.. thank you for using #BAML and dealing with it..." 😂 Boundary (YC W23)

1k ⭐️!

Hitting near 100% success with smaller LLM models like Microsoft's Phi3 or Google's Gemma2 used to be hard. Using BAML with Ollama apparently solves it. No Fine-tuning. Also, who woulda guessed Barry would have spent a few hours Monday morning writing in our made up programming language... Boundary (YC W23)

🔍 BAML for structured extraction from LLMs I recently delved into a fascinating tool called BAML (Basically A Made-up Language, it's a real name), a solution for structured data extraction. BAML not only simplifies the coding process but also eliminates the frequent headache of language-specific discrepancies. Whether you’re coding in Python or TypeScript, BAML maintains consistent behavior, backed by robust type safety. For a deeper look into BAML, you can check out my blog post: https://lnkd.in/dnt2N-HT Why Use BAML? Utilizing BAML for structured data extraction from LLMs not only enhances code efficiency but also accelerates development cycles and improves outcomes in deploying AI-driven applications. By abstracting complex underlying interactions with language models, developers can focus more on application logic and less on the nuances of specific LLM APIs. Boundary (YC W23) Also --> https://lnkd.in/d5mu_f9J

🛠️ Libraries to get structured output from LLMs Nice blog post by Sam Lijin from Boundary ML about the different techniques to get structured generation with LLMs. As a disclaimer, Boundary LM is also the creator of the BAML framework. I've personally tested JSONformer, Guidance, LMQL, and Outlines (in this order). I would definitely recommend Outlines and LMQL over the others. In any case, I would strongly suggest choosing a framework where you have full control over the prompt. We saw libraries sneakily asking GPT to output JSON with prompt engineering in the past... :) 📝 Article: https://lnkd.in/eV5M9j33