Of course, I say that right after you've already pressed it.

Give me one second.

I didn't put this in my prompter.

This is the Daily Tech News for Thursday, April 30th.

We're all out of April.

2026.

We tell you what you need to know, give you the important context, and help each other understand.

Today, why goblins infested open AI's codex, how it got rid of them, and what it means for understanding model training.

I'm Tom Merritt.

I'm with Twett now.

Let's start with goblins.

So last week, OpenAI released source code on GitHub for the Codex Command Line interface, or CLI.

This revealed that the system prompt for the CLI,

system prompt is something that reigns in its behaviors, no matter what you type in.

So it included instructions, and I quote, never talk about goblins, gremlins, raccoons, trolls, ogers, pigeons, or other animals or creatures, unless it is absolutely and

unambiguously relevant to the user's query. That prohibition is repeated twice in the 5,000 words of

the system prompt. Other elements of the system prompt are probably more understandable, like never use destructive

commands like Git reset hard or get checkout unless the user is clearly asked for that operation.

That's a good one to have in there.

Also, I think a lot of people will like that there is a system prompt that says not to use emojis or m-dashes unless explicitly instructed.

I think we all like that one too. But up until now, I don't think there's been a whole lot of discussion. There has been some, but not mainstream discussion of chatbot goblin talk or raccoons.

Certainly not as much as there has been about M-dashes.

...