TypeSafe AI's Jev turns unstructured input into typed decisions and probabilities. We rebuilt the API with an open model, reached 113ms median latency on an M4 Mac, and used it to play Doom.
nice! like HTML tags. I remember seeing this back with meta Llama 2. the <|start_header_id|>assistant<|end_header_id|> style. And [TOOL_CALLS] ... [/TOOL_CALLS] style with Mistral
Yes! You saw it immediately. Article was written by Claude. It has a distinct cadence that matches all the way through writing. It's the same in the CLI claude code. I'll make sure the upcoming lesson will be less ai generated text. I usually like to focus on writing the most read parts. Like the intro and core focus of the article
You a are right as The article doesn't mention KV cache. Also, yes the writing is slop, and does not mention the LLM special tokens. Only discusses the use through the OpenAI-style JSON wrapper that allows you to define the schema of a tool call in JSON. For most of the audience, they are looking to build AI agents, and the high-level tool calling interface is what they would be using
Yes this is true, you even got the model correct. I relied too much on Claude this time. I'm starting to work on the next AI agent lesson this week. I'll do better on the upcoming lesson. The next lesson is A2A focused
Early to be using this technology for coding in the browser. Our consumer devices are underpowered. It is fun to see how far we can take it today. Here is the walkthrough for running LLMs client-side with WebLLM using WebGPU. Initializing and using @mlc-ai/web-llm with model downloads, caching, and progress tracking. Using the streaming completions sent to pre element with zero network calls after it is loaded. Also handling WebGPU memory limits. Includes code blocks and an end-to-end video walkthrough.
that is the strongest point I've heard. who would this be for? Replit, Bolt.new and Lovable are mostly-in browser vibe code engines. Target audience would be for a free local version alternatives. Though maybe that isn't something that anyone needs.
The original intent for this project is for learning how to build agents, and finding what are the limits in the browser. Especially now that we have WebLLM. Maybe we can find a use for it with local agents
You are right, there's not enough time to filter a bad hire. Not enough time to fully vet anyone during an interview. We can work a challenge together to see how you work collaboratively
Interviews cover collaboration compatibility. Not enough time to quiz you on tech knowledge. Leetcode-style interview represents your ability to work a problem collaboratively with a team
reply