AI is good for expanding your perspective, giving you ideas, but you can't delegate control to it.
A Harvard study found that 15,000 simulations across seven major AI models all produced the same trendy—but not necessarily correct—business advice. The speaker argues this finding is partially valid but misses the point: AI is only as useful as the questions you ask it, and users must critically engage with it rather than blindly delegate decisions to it.
Summary
The video opens by referencing a Harvard study that ran 15,000 simulations across seven frontier AI models—including GPT-5, Claude, Gemini, and Grok—and found that all of them converged on the same business advice. Critically, the study suggests this consensus answer was not the correct one, but rather the trendy one, implying that AI models are biased toward popular or conventional thinking rather than genuinely optimal solutions.
The speaker takes a nuanced position on the study's conclusions, describing them as both correct and incorrect simultaneously. On one hand, they acknowledge that AI can absolutely be made to look foolish depending on how it is prompted. On the other hand, they push back against the implied takeaway that AI is simply unreliable for business decision-making.
The speaker's core argument is about prompt quality and user responsibility. They explicitly state they would never ask AI to simply 'solve a business question,' framing that as an ineffective and naive approach. Instead, the proper use of AI requires critical engagement—challenging its outputs, probing its reasoning, and treating it as a thinking partner rather than an authority.
The video concludes with a clear principle: AI is valuable for expanding one's perspective and generating ideas, but control and final judgment must remain with the human. Delegating decision-making authority to AI is presented as a fundamental misuse of the technology.
Key Insights
- Harvard researchers ran 15,000 simulations across seven frontier AI models and found they all clustered around the same business answer—not the correct one, but the trendy one.
- The speaker argues the Harvard study's findings are simultaneously correct and incorrect, suggesting the conclusion depends heavily on how AI is being used.
- The speaker claims that any AI can be made to sound idiotic if you ask it the wrong questions, implying prompt quality is the determining factor in output quality.
- The speaker explicitly states they would never ask AI to simply 'solve a business question,' arguing that users must critically challenge AI rather than accept its answers passively.
- The speaker concludes that AI's proper role is expanding perspective and generating ideas, but that delegating control or decision-making to AI is a fundamental mistake.
Topics
Transcript
[0:00] Harvard allegedly just proved that every AI gives the same [music] bad business advice. Researchers run 15,000 simulations across seven frontier models, GPT-5, Claude, Gemini, Grok, and they all clustered around exactly the same answer. Not the right one, the trendy one. I think this is both correct and incorrect at the same time. I think [music] that you could ask AI any question and make it sound idiotic, 100%, [music] but you've got to know the right questions to ask it. I would never say to AI, "Solve this business question [0:31] for me." Right? You have to critically challenge it. The key thing is it's good for [music] expanding your perspective, giving you ideas, but you can't…
Full transcript available for MurmurCast members
Sign Up to AccessMore from Jack Roberts
100 hours of Hermes Agent lessons in 23 minutes
Jack walks through advanced features of Hermes Agent, an AI personal assistant, covering memory systems, background tasks, scheduled cron jobs, model switching, and integration with tools like Obsidian, GitHub, and various AI models. The video aims to help users unlock capabilities beyond basic chatbot usage. Key themes include connecting external memory systems, delegating tasks to specialized AI models, and building a persistent, context-aware personal assistant.
Claude Code = $10,000 Beautiful Slides
Jack demonstrates a GitHub-based system using Claude Code to generate professional slide decks from any website URL by codifying 20 universal design principles. The system uses Firecrawl to extract brand DNA from websites and can integrate AI-generated images via APIs like Krea.ai. The result is polished, brand-consistent HTML presentations created in minutes with minimal input.
The most valuable thing I got from working at McDonald's
The speaker reflects on their first job at McDonald's at age 17, earning £5 an hour. The most valuable takeaway was the realization that they never wanted to work such a job again, reinforced by calculating how long it would take to earn a million pounds at that wage.
Claude Code Memory System = CHEAT CODE
Jack Roberts presents a three-tier Claude memory system designed to give AI tools persistent context across all platforms and sessions. The system consists of short-term identity memory, mid-term project memory via structured folders and claude.md files, and long-term memory using either Obsidian or Pinecone for archiving conversations and expert knowledge. The goal is to eliminate the 'amnesia' problem where AI loses context between chats.
ChatGPT ads aren't actually a bad thing. Here's when I want them.
The speaker argues that ads in ChatGPT are not inherently bad, particularly when users are actively seeking product recommendations. They distinguish this as 'pull advertising,' where the user initiates the request and ChatGPT synthesizes results based on provided context.