When are we going to start talking about LoRA routers instead of model routers? LoRAs are like a pair of prescription glasses you put on a model to tweak its weights. LoRAs are better than distilled or finetuned models because they are super tiny compared to the model itself.
Thinking about making a zine or two on AI topics I am passionate about, starting with local models, then maybe evals and MCP. Could opensource the files so people and event organizers could print them themselves. Thoughts?
The theme of Typescript AI this year was "factories," a far cry from AIE's "slow down." As agents occupy more and more of the Developers' Journey, IRL events that offer human connection are more precious than ever, offering rare moments of connection and collaboration. Thank you, Mastra team!
I'm pumped to be at Google I/O Connect starting tomorrow! Anyone want to connect about local models, the agentic web, and/or observability?
"How do you build an app with AI?" I doodle a lot. And talk a lot with people and models about what the best architectural approach is. I put the doodles and the ideas into a backlog of issues. They get grouped by feature. When they hit critical mass, I work with a group of agents overnight
I just met a 17 year old who started learning about LLMs three years ago and just built their own model for detecting toxicity and grooming. I'm pretty sure I just had a brush with greatness
Returning to well with this one. I often say "people don't read, they look." And it's true that we learn to understand vision, touch, hearing, all our senses, before words. Makes me think about the limitations of AI
Out here shipping at @OpenAI Build Week🔥 What you up to?
Sourcing 3D modelers in the era of AI agents is full of false positives.
TypeScript AI is a lovely conf. Thank you, Mastra
