i'm actually curious if anyone has explored syntax highlighting with ML. doesn't have to be an LLM, but something that actually highlights what's interesting, or even strange, about a code snippet.

I did somewhen toy with the idea of comparing how a line of code is written to what an LLM-based next-token-prediction would emit for the same line. Then it might be possible to highlight the pieces of code that most differ from what an LLM would have written.

The base form isn't too useful by itself, but I expect if one would "seed" the LLM with a prompt about focusing specifically on a particular class of errors, it might be possible to find violations by seeing where the code completely deviates from what an LLM would have written.

Never got around to doing anything with this, but I expect a prototype should take no more than a day with some agent-assisted coding ;) Tell me how it went if you try!

i assume this is largely about the cybersecurity blitz, to ensure hot-path code is secure?

How is this different from tool calls? Or rather, MCP?

It's likely just tool calls, so it might ride over MCP:

> It is also model-agnostic, and any agent harness can access it using standard protocols, such as the Model Context Protocol.

I think pinning WebMCP to the browser was the wrong choice. To list or call tools, you have to have access to the browser environment. "Declarative" WebMCP allows you to read the HTML, but to call tools you still need the JS execution env.

This limits the utility of WebMCP to in-browser agents. In practice, what's the use case? Computer Use can work, but at that point you can circumvent WebMCP entirely.

IMO the better option would be to unify MCP with a standard protocol, like HTTP. Right now, you have to rely on the MCP registry for discovering MCP servers, but wouldn't it make more sense to have this natively in HTTP?

How long have "large scale RAG systems" really existed in the first place? I'm always surprised at this, given how new all this really is, relatively speaking.

my personal experience - I have been involved with such projects for the last 2 years. interestingly enough, a lot of times such initiatives didn't take off because people/stakeholders were overcomplicating things and wanting to use semantic search for everything - without having a minimum knowledge of chunking strategies, pros/cons etc

how does agents.md work for subagents and swarms? is it actually that useful?

yes, in the tools I use, each (sub)agent will also load the same files (automatically, root is always loaded, a nested AGENTS.md is read if the dir or a peer file is touched)

yes, they are useful, mainly in that they shorten the context gathering phase and can call out gotchyas, keep it minimal

A good one does, yeah.

A bad one is noticeably harmful.

I recall there was another YC startup that was working on Mac-specific ML optimizations for local inference (and perhaps fine-tuning).

I wonder if their work is related?

RunAnywhere or Conifer?

Hm is the intent to one day replace the CPU?

The goal is to use similar abstractions and code across both the CPU and GPU where it makes sense.

i've asked llms to write a presentation for me on a given topic. for whatever reason, that finds the hidden layer responsible for explaining things well.

outside the slop, i liked this post that was linked on innodb locking: https://jahfer.com/posts/innodb-locks/