Hype Check · Ras Mic

Worth your time?

Name: Is "I made AI play Among Us (Claude LIED LOL)" worth watching? Hype 29/100 | Clip'd
Item: I made AI play Among Us (Claude LIED LOL)
Rating: 71
Author: Clip'd

I made AI play Among Us (Claude LIED LOL) by Ras Mic

Watch on YouTube

Hype score · lower is better

29/100

Useful but promotional

This video mostly demonstrates a fun, reproducible Among Us AI demo and explains limits, so it teaches and entertains more than it sells.

Substance60

Proof50

Pitch pressure25

Claim inflation10

The 15-second version

Post it, send it, settle the argument.

Download clip

Receipts

Why this score

Scored from the transcript, summary, and output signals.

Title deliveredTitle promises AI playing Among Us (and Claude lying); the video runs the AI simulation, reveals roles, and shows deceptive behavior, so it delivers.

Skip the video

The short version

Kept by Clip'd

The creator used Anthropic's /goal feature to build a quick Among Us simulation where different LLMs play crewmates and an impostor. The video shows the game running (simulated movement, kills, and meetings), explains how /goal produced the system with one prompt, and evaluates when /goal is useful, mainly for prototyping and experiments, not production. The build used multiple models and 11 Labs for voice, and the creator argues /goal is powerful for idea exploration but limited by assumptions, token cost, and lack of handoff.

Keep the lesson

What it teaches

Use 'goal-directed' agent modes like /goal for rapid prototyping and ideation when you want the model to autonomously iterate until a completion condition is met. Apply this when you need a fast proof of concept, a playable demo, or to explore emergent behaviors across multiple agents. Do not use it for production feat

Jump straight there

Key moments

1:50Model roster revealed: lists the six models used (Claude Opus 47, Sonic 47, GPT4, Gemini 3 Pro, Claude Opus 46 Fast, Gemini Fast Latest), which frames the rest of the experiment.
2:17First kill by Opus46Fast: shows the agent behavior in a tactical moment, illustrating that models will fake tasks and opportunistically kill.
4:53Opus46Fast kills again and almost runs the table: a turning point where the impostor nearly avoids any meetings and controls the game flow.
5:41Meeting begins with fragmented responses: demonstrates models' limited conversational reliability during short, adversarial interrogation.
7:23Vote and impostor win: the simulation ends with the crewmates misvoting, proving emergent strategic outcomes can arise from the setup.
7:36Technical note, 11 Labs API used for voices: confirms the audio layer was synthesized via 11 Labs, an important integration detail.

Lines worth keeping

Quotes

It is a lot cooler than it is useful.0:51
I don't think you can build anything useful with this because the AI is going to make a lot of assumptions.10:28
Ralph Wiggum is officially dead. It's now /goal.11:05

Clip'd

Skip the video. Keep the lesson.

Score your next video free

The Hype Board