Neural Toons Original

LLMs Can't Bluff: Llama 3.1 70B Loses 23.2% More as Fascist in Secret Hitler

25 May 2026 · The Neural Feed
LLMs Can't Bluff: Llama 3.1 70B Loses 23.2% More as Fascist in Secret Hitler – Neural Toons comic

“Because when your AI can't manage a bluff in a party game, it's definitely ready for production.”

The story behind the toon

What happened: Because when your AI can't manage a bluff in a party game, it's definitely ready for production.

This Neural Toon is based on the story “Evaluating Large Language Models in a Complex Hidden Role Game”.

Read our summary of the story · original source

Comic transcript

Panel 1 – Jordan: Our AI bot needs to bluff customers. This is dangerous.

Panel 1 – Taylor: Deadline's Friday. Ship it and iterate.

Panel 2 – Jordan: Secret Hitler study: Llama 3.1 loses 23.2% more as fascist.

Panel 2 – Taylor: That's a game, not our enterprise bot.

Panel 3 – Jordan: Rule-based agents beat LLMs 86.7% to 59.7% in voting.

Panel 3 – Taylor: But our bot needs to sound like a human.

Panel 4 – Jordan: Chain-of-thought made Fascist performance even worse.

Panel 4 – Taylor: Wait, more thinking leads to worse lying?

More Neural Toons · Neural Snaps · The Neural Feed