Learning/building with @CopilotKit and @langchain reminds me of learning Kubernetes in 2018. There’s a bit of a learning curve, lots of people think it’s overly complex, but it feels like learning the core abstractions that will be used for all future apps
Commanders need clarity. @15thMEUOfficial used our platform in the Pacific to cut through the noise and get a real-time picture of their impact. Mission-ready AI at work!
https://t.co/IicJL5BZaU
Today Vannevar Labs announced the transition of the company’s prime contract with @DIU_x from a prototype contract to a production scale contract. This production agreement will field capabilities developed under the prototype agreement for exquisite sensing capabilities that enable DoD teams to identify adversary activity, manage their forces’ signature, deliver effects, and assess the impact of those effects in strategic competition.
Learn more here: https://t.co/zL3hzGNAyp
(1/2) When LLMs first came out there was a lot of alpha in being a great frontend engineer because you could quickly spin up interesting demos using out-of-the-box LLMs. As LLM-based agents become mainstream there will be a lot of alpha in being a great backend engineer.
New post on a technical approach I’ve been experimenting with that I’m calling GAL - Generate, Attribute, Localize - getting LLMs to pinpoint what parts of the original source data influenced their output in a retrieval augmented generation setup. https://t.co/a8PfWysX0K 🧵
I show how to make this work on any data type from text to audio to PDFs.
This aligns closely with new open-source work by @jxnlco and commercial work like @cohere’s Coral and @magicaltome's Reference feature.
(1/5) I'm open-sourcing a new project today called MirrorGPT aimed at making personalized Agents/LLMs and solving the individual alignment problem.
Mirror provides the structure, and some early implementations, of 3 core functions for building personalized Agents:
@SurfingPigment I'm focused on the AI-core and won't get to Vercel/Replit for a bit but would love to see a contribution in that area!
I have been looking into a Whisper integration today and hope to have that one soon