CharlieKey takeaways from your favorite podcasts
← 3 November 2026The AI Daily Brief

How to Deslopify Your AI

AI slop is work where the person skipped the thinking and shifted the cost to the reader; avoiding it means thinking first, setting review points, and using AI to sharpen your own view rather than replace it.

Watch on YouTube ↗

Charlie's summary

  1. Slop is work where the sender skipped the thinking and pushed the cost onto the receiver

    Nufar defines slop as work where the person who should have done the thinking skipped it, partly or fully, so the cost moves to whoever receives it. She cites a Stanford and BetterUp study: about 40% of US desk workers got slop in the past month, costing close to two hours. Another survey found almost 70% shipped work unread. The standard is the least effort for the reader.

    ▶ 3:34
  2. Models produce slop by design, and switching vendors won't fix it

    Nufar explains that a language model predicts the most likely answer, so a generic prompt returns the internet's average. Tuning adds a typicality bias because raters reward the familiar. A NeurIPS best paper found models from different companies converge on the same answers. Her test: the same LinkedIn prompt in ChatGPT, Claude and Gemini gave near-identical results, but adding her own point of view gave a much better draft.

    ▶ 6:39
  3. Humans tolerate slop because we are cognitive misers who outsource thinking

    Nufar says the brain saves energy where it can, like pilots' automation complacency: when the machine is usually right, you stop checking. Research shows more trust in AI means less critical thinking. Slop comes from easy-looking output under pressure, not knowing the craft, and asking the tool what it thinks before deciding what you think. Her crusade is “recovering the lost art of thinking.”

    ▶ 9:12
  4. Test for slop: could you defend every sentence, and would you put your name on it in capitals?

    Nufar's detection questions: Could the sender defend each sentence without rereading? Did they read all of it? Is there anything only this person could have contributed, like a point of view, a decision or real data? Does the format save the reader time? Her favorite is whether you'd put your name in all capital letters on it. AI detectors and humanizers don't help much: they measure style, not originality, and humanized text is still flagged about three quarters of the time.

    ▶ 10:44

Give your email to unlock the rest of the key points — for every episode.

Do more with Charlie

Free — sign in with your email, no password.