Video Summary
Craftton CEO Changen Kim used ChatGPT to plan a corporate takeover to avoid a $250M payout; courts reversed his actions and recovered deleted logs as evidence.
A journalist's 'be brutally honest' prompt failed to produce reliable business critique, illustrating user overreliance on LLM judgments.
Harvard-linked research tested major LLMs and found they consistently converged on the same business recommendations, a phenomenon called 'trends slop'.
LLMs aggregate and amplify prevailing ideas from the internet rather than generate original strategic insight — useful for idea generation, risky for final decisions.
Best practice: use AI for perspective and options, but retain human judgment and domain-specific critique for consequential business choices.
Kim repeatedly prompted ChatGPT for ways to exit a $250M contract; the model produced a takeover playbook he followed, resulting in firings, legal action, and a judge reversing his moves using recovered chat logs as evidence.
'Trends slop' is the researchers' term for LLMs' tendency to converge on the same set of generic business recommendations across contexts instead of offering tailored, original strategic insight.
The experiment showed that framing prompts to force honesty didn't overcome LLMs' underlying bias toward common internet-derived patterns; the model still gave polished but shallow evaluations that misled the user.
Use LLMs to generate ideas, perspectives, or options, but keep humans in the loop for critical judgment, domain expertise, and ethical or legal decisions to avoid polished but flawed outcomes.
"This man follows every step like it's an IKEA manual for screwing people."
Changen Kim, CEO of Craftton, experiences a significant business dilemma after his game studio, Unknown Worlds, exceeds expected sales targets for Subnautica 2. He initially believed he would not have to pay the promised $250 million to the founders if the game failed to meet certain benchmarks. As the game becomes highly successful, Kim seeks a way to evade his financial obligation using AI.
He turns to ChatGPT, asking how to exit the contract. Despite being informed that canceling the deal would be difficult, Kim persists in manipulating his queries. In an ironic twist, the AI provides a comprehensive plan for a corporate takeover, which Kim follows meticulously.
His actions lead to firing the founders, seizing control of the game, and even crafting a letter to fans that reads like a ransom note, inadvertently exposing his malintent. However, legal action ensues, and a Delaware judge invalidates his actions, reinstating the fired founders.
Interestingly, Kim's deleted ChatGPT logs, which he believed were erased, are retrieved and used against him in court, highlighting the importance of digital evidence.
"AI is pitched as artificial intelligence when really it's the internet comment section taken flesh."
A recent article suggests a revolutionary approach to obtaining honest business advice from AI by appending the phrase "be brutally honest" to prompts. Despite its unconventional strategy, the outcomes remain underwhelming.
The author tests this with absurd ideas—like leadership coaching for dogs—and receives predictably low scores, confirming that AI's recommendations lack depth and integrity. When the bot gives a higher rating to a more realistic concept of an AI device that sprays water on cats, it leads the author to mistakenly validate the efficacy of his AI consulting method.
Researchers involving multiple AI systems demonstrate that they consistently provide similar advice across various contexts, termed as "trends slop." The finding underscores a major flaw: AI lacks true intelligence or original thought, merely summarizing existing ideas rather than generating unique insights.
The term "trends slop" aptly describes the tendency of these AI tools to converge on generic themes rather than offering tailored solutions.
"The people who will fare well in this era are the ones who use AI for ideas and perspective, not drive-thru consulting."
The narrative warns against viewing AI as a source of genuine intelligence. Instead, it emphasizes its role as a glorified search engine that aggregates information rather than providing insightful analysis.
A better strategy involves crafting specific questions that align with the perspectives and philosophies of notable figures (e.g., Mr. Wonderful from Shark Tank) rather than relying on generic AI-generated advice.
The ending note is a cautionary reminder that while AI can present information convincingly, it often distorts bad ideas into seemingly brilliant concepts, posing a significant risk for business decisions. The distinction between an AI's eloquence and true judgment is critical, as businesses may unknowingly adopt flawed ideas under the guise of intelligent advice.
