4 min read AI

Why LLMs are like a pub

Stop treating Generative AI like a divine Oracle. It isn’t. It’s a pub.

A massive, deafeningly loud, (almost) infinite pub.

If you walk into a pub and shout, “How do I fix a broken supply chain?”, you are accessing a form of collective intelligence. You will get a lot of noise. You will get the loud guy at the bar who thinks he knows everything because he once read a blog post.

But, if you listen to the aggregate, the low hum of conversation; the consensus of the room, you get the Median Position. 

And I think it’s probably OK to say:

The Median Position is actually useful.

Most corporate work doesn’t require Einstein. It requires “Competent Consensus.”

The “Blank Page” Reality

Let’s look at the Internal Comms Lead.

They have to write a policy update on the new hybrid working arrangements. It’s dull, it’s sensitive, and they are staring at a blinking cursor.

  • Scenario A (The Old Way): They stare at the wall for two hours, terrified of getting the tone wrong, write three drafts, delete them all, and go for a panic-coffee.

  • Scenario B (The Pub Method): They ask the AI. Effectively, they walk into the pub and ask, “What’s a standard, empathetic, yet firm way to tell people to come back to the office?”

The AI gives them the Median Position. It gives them the average of every well-written, corporate-standard policy in its training data.

Is it winning a Pulitzer? No. Is it a solid, B+ draft that they can edit in 20 minutes? Yes.

The “Pub” has solved the inertia. The Comms Lead is no longer blocked; they are editing.

The Risk: Mistaking the Median for the Expert

The danger arises when you ask the Pub for something that requires the outlier, not the average. When the Median gets it wrong, we call it a “hallucination” — but as I’ve explored before, that tendency mirrors our own cognitive patterns.

If your General Counsel asks the AI to draft a complex IP indemnification clause for a bespoke software deal, and they accept the “Median Position,” they are effectively using a contract that “looks about right” to the average person.

That’s not good enough. That’s negligence.

A Solution: A People, Process, Technology Checklist

You need a strategy for when to trust the Crowd (The Median) and when to find the Expert.

1. People (The Editor)

  • Acceptable: Using the Median to generate volume, drafts, summaries, and standard code.

  • Not Acceptable: Allowing the Median to make the final decision.

  • The shift: Your people stop being creators from scratch and start being rigorous editors.

2. Process (The Filter)

  • Listen: What is the AI actually giving you? Is it generic waffle or specific insight?

  • Adapt: You must force the tool out of the “pub main room” and into the “private booth.” This is Prompt Engineering. It’s the difference between shouting at the bar and interviewing a specialist.

  • Enable: Teach your teams to recognise “Average” vs. “Good.”

3. Technology (The Tool, Not the Talent)

  • Remember, the technology is a pattern matcher. It craves the middle. You have to drag it to the edge if you want brilliance.

The Artisan’s View

Collective intelligence is powerful because it is safe. It smooths out the rough edges of individual stupidity. But it also sands down the sharp points of individual genius.

And, like when you’re at the pub: have your critical thinking cranked up and on alert.

Use the Pub to clear the backlog of average work. But don’t let the noise of the crowd drown out the actual experts in your business.

Where do you draw the line? Are you letting the “Median” drive your strategy, or just your emails? 

Article assisted by AI.