Brian Maucere

And I'm alongside with you, Anne.

I'm a normie.

I am not a coder.

codex and i see why i've just recently sort of gotten um i wouldn't say i'm deep into it by any means but i've certainly poked around more than i had any time previously and this is just yeah so this the there's a weave off of that yeah go for it what microsoft has in their new co-pilot multi-model

architecture.

Okay, so not multimodal, that's different, multimodal, and we're talking about Claude having now a plugin for Codex, which lets you call Codex from within a Claude code session.

And so you're using multiple models now.

And there's a recent post by Andrej Karpathy that I didn't read that I want to try to find where he says, this is the case.

This is the way it should be going forward.

You should be having adversarial checks and balances across different models to avoid the mistakes that models can make.

And boy, I had a long, long library of mistakes that models have made for me.

Here's what Microsoft did.

They just made their co-pilot researcher, that's a feature of co-pilot, that uses two models in sequence.

OpenAI drafts the research, and then Claude critiques the outputs.

And it's, you know, a model council mode that's also associated with this new multi-model researcher feature.

And that mode runs side by side and flags where they agree, disagree.

So not only does it pit two models against each other to do the research, but then it has a counselor that says, okay,

Here's the summary of what the differences are between those two.

I think that's brilliant.

This multi-model orchestration is what you're talking about, Brian.