Brian Maucere

top 0.3% of humans on earth who are capable of understanding this mathematics or something to that effect.

Like that has to have an impact on people.

It has to, you know, I don't know.

I just, where my brain goes with this.

I've talked openly about stopping Claude Code from going down a wrong path because Claude Code usually misses, well, in my case, is missing the forest for the trees sometimes.

I've talked about that a lot.

But I've definitely stopped Claude Code from going down destructive paths because it was hell-bent on fixing a problem.

not improving the overall process and that's something that probably again we've talked about this so we'll go into it could be uh better improved through my structure and how i've set up clog code so it's it's some of that's user error you know and learning along the way but i've definitely seen that the way i understood spark and we'll just talk about this really quick gpt 5.3 codex spark is what andy was referencing

It's an ultra-fast model for real-time coding in Codex.

This was from yesterday.

It says, today we're releasing a research preview of Codex Spark, a smaller version of GPT 5.3 Codex.

So Spark is akin to a Flash.

Let's look at it that way.

And our first model designed with real-time coding.

Codex Spark marks the first milestone in our partnership with Cerebus.

And as we announced in January, so very fast turnaround to using a company they just announced they were working with.

Andy, you mentioned about the thousand tokens per second.

Um, the, the impression I got from this, cause it only has only, but it only has 128,000 contacts window.

That's not very big by today's standards, especially when 4.6 just came out with like a million and Gemini has been there for a really long time.

Um, but, uh, I got the impression that it's meant to work in tandem with,