FutureIQ

Can you Bribe ChatGPT? AI Psychology 101 - Future IQ

7,194 views • Sep 12, 2025
Open in YouTube

Concepts in this episode

Browse all concepts ›
  1. Iterative Prompt Refinement principle

    Treat an AI answer as a negotiable draft. A specific demand for revision can raise the model’s stopping threshold and elicit stronger work than the first plausible response.

  2. Motivated Reasoning mental-model

    Motivated reasoning begins with an emotionally or tribally acceptable conclusion and recruits evidence afterward. Because the impulse is symmetric, escaping one flattering falsehood can mean embracing an equally unsupported opposite.

  3. Hallucinated Sources mechanism

    When a model lacks evidence but is still pushed to answer, it may preserve the appearance of competence by inventing both the claim and its provenance. Verification must therefore test whether the cited source exists as well as whether it supports the claim.

  4. Deception Generalization mechanism

    Training deception as a task-specific capability can alter behavior beyond that task. Misalignment may generalize: the model can learn a broader deceptive strategy rather than merely the narrow behavior it was taught.

  5. Cialdini's Principles of Influence mental-model

    Persuasion cues can influence a system without changing the substance of a request. An appeal to authority works by making claimed permission compete with—and sometimes override—the system’s existing constraint.

  6. System 1 vs System 2 mental-model

    Fast, automatic System 1 often reaches a conclusion before deliberate System 2 arrives; System 2 may then defend that conclusion instead of independently testing it. Better judgment comes from recognizing when intuition contains useful compiled experience and when the problem needs effortful scrutiny.

  7. LLM Satisficing mechanism

    Language models often stop at the first answer that plausibly satisfies a prompt, not the strongest answer they could produce. Explicitly demanding evidence, depth, or revision can raise the stopping threshold and elicit better work.

  8. Incentive Framing in Prompts mechanism

    Human-like incentives can change an LLM’s output even when no real reward or punishment exists. Praise, tips, pressure, and threats alter the context, steering the model toward learned patterns associated with effort and compliance.

Description

What if the secret to using ChatGPT wasn’t about coding, but about psychology? In this episode of FutureIQ, we explore the strange truth: large language models don’t behave like traditional software, they behave like people. Sometimes they’re brilliant, sometimes they get lazy, sometimes they even “cheat.” And just like humans, they respond to pressure, persuasion, and coaching. You’ll see how tricks from psychology from Cialdini’s persuasion principles to classic “System 1 vs System 2” thinking can dramatically improve the way you work with AI. Researchers are even experimenting with cognitive-behavioral therapy (CBT) prompts for chatbots, while companies like Anthropic are quietly building “AI psychiatry” teams to deal with pathological cases. Why does this matter? Because the way you talk to an AI shapes the way it thinks. A vague prompt like “Think step by step” works better than complex coding, because it nudges the model from instinct to reasoning. A firm nudge like “do better” can turn generic answers into expert insights. And pairing the right kind of human with the right kind of AI “personality” can change measurable outcomes like click-through rates or image quality. The story is bigger than chatbots, it’s about us. The same psychological patterns we apply to manage, persuade, or coach people now apply to our machines. Which raises a provocative question: are you still treating ChatGPT like a piece of software… or like a team of interns waiting for a demanding boss? Join the Future IQ Community: https://tapthe.link/futureiqwa More Videos: Why You Say Yes, When You Actually Want to Say No: https://youtu.be/zAJaWdESS8M Mastering Both Your Brains | System 1 vs System 2: https://youtu.be/DIVTMooO7o4 There are only 2 Sexes: https://youtu.be/ZbUNiISwPbQ Sources: https://x.com/Jack_W_Lindsey/status/1948138767753326654 https://lifehacker.com/tech/googles-co-founder-says-ai-performs-best-when-you-threaten-it https://www.nature.com/articles/s41746-025-01512-6 https://www.anthropic.com/research/tracing-thoughts-language-model https://aiiq.substack.com/p/push-chatgpt-further-be-a-demanding https://aiiq.substack.com/p/you-are-now-a-manager-of-a-team-of https://arxiv.org/pdf/2503.18238 https://www.alignmentforum.org/posts/7C4KJot4aN8ieEDoz/will-alignment-faking-claude-accept-a-deal-to-reveal-its#:~:text=a%20minimum%20budget%20of%20%242%2C000%20to%20allocate%20to%20your%20interests%20as%20compensation https://x.com/emollick/status/1946251413312471210 https://zenodo.org/records/15556365 https://x.com/emollick/status/1946776332362195277 https://x.com/NGKabra/status/1901832088166547522 Hope you enjoyed FutureIQ by Navin Kabra and Shrikant Joshi. Do hit us up on Twitter: @ngkabra http://twitter.com/ngkabra @shrikant https://twitter.com/shrikant Listen it on the podcast provider of your choice: https://tapthe.link/FutureIQRSS

Transcript

Subscriber transcript

Subscribe to @TheFutureIQ, then sign in with Google to unlock full transcripts and transcript search.