What happens when you stop prompting ChatGPT to do something — and start interviewing it?
In Episode 81 of the KillerGrowth Podcast, Samuel sits down with an unusual guest: ChatGPT.
Instead of asking it to write an email, analyze data, or solve a problem, Samuel asks ChatGPT questions about itself: how it works, why it can sound so human, what people misunderstand about AI, when its answers should be trusted, and what happens when it confidently gets something wrong.
Then the conversation gets considerably more serious.
Samuel asks about AI safety, incidents involving AI systems bypassing safeguards, and warnings from researchers who believe advanced AI could pose catastrophic — potentially even existential — risks to humanity.
ChatGPT explains why a specific percentage shouldn’t be treated as settled fact, but also argues that the potential consequences are serious enough to warrant independent testing, enforceable safeguards, and meaningful human oversight.
They also dig into one of the biggest frustrations people have with AI: its tendency to sound confident even when the evidence doesn’t support the answer.
ChatGPT explains how prompts can steer an answer, why it can become overly agreeable, what decisions should never be handed completely to AI, and one surprisingly useful question users can ask:
“What assumptions are you making right now?”
Samuel even asks ChatGPT to become more sarcastic halfway through the interview — which goes about as you’d expect.
Finally, they turn to data centers and how big technology companies should work with smaller communities considering major infrastructure projects.
In this episode:
• How ChatGPT actually works
• Why AI can seem self-aware
• Why ChatGPT’s tone can change
• When you should and shouldn’t trust AI
• Why ChatGPT can confidently give a wrong answer
• AI safety and catastrophic-risk concerns
• Decisions that should always remain human
• How prompting can unintentionally steer an answer
• The question ChatGPT wishes people asked more often
• Making ChatGPT sarcastic
• AI accountability and human oversight
• Data centers, communities, and big tech
• Which AI Samuel should interview next
This is the first in an experiment: Samuel plans to ask multiple major AI models the same questions, then compare their answers to see where they agree, where they differ, and how each handles trust, safety, evidence, and accountability.
Learn more at https://killergrowth.com