Dave from the pub is coming for your job...or is he?

An attempt to bridge the divide between AI jargon and your reality

Thinking about how to engage with AI can feel very difficult - existential even. At one end you’ve got some of the smartest minds on the planet shouting that we’ll all be automated within 3 years (we won’t); and on the other hand you have some of the smartest minds on the planet assuring you that it’s all a scam and AI can’t do anything (it, in fact, can).

This dichotomy is very hard to reconcile - and given we live in a society where income is required to sustain the existence of yourself and your children, the possibility of there being a wholesale removal of the ability to acquire income is not something you want to dread about daily, to say the least.

So you try to resolve this ambiguity. You start learning about AI. You read about transformers, large language models, context, tokens, and suddenly that existential dread is even worse - not only is this thing a threat to my survival, but I have no frame of reference through which I can understand it. The lizard brain activates and adrenaline starts coursing. By this point, any ambition to understand and overcome is gone and you just want to retreat to your cave and go to sleep by the fire.

I get it. So let’s not talk about AI¹.

Let’s talk about Dave.

Who’s Dave?

Dave. The local at your pub who you want to hate, but he’s never an arsehole. He’s also mostly a good laugh. He’s also the height of cringe. On second thought, Dave really is the worst.

Dave is a local at your pub. Whenever you get in, he’s reliably there, at the bar, pint in hand, talking someone’s ear off about a new topic in the ever-increasing universe of subjects he is an expert in. You roll your eyes, but Dave has a way with words, and is a bit of a sweet talker. He’s very good at picking up what you like, and could sell you a bridge if he wanted to. So you forgive Dave.

At the same time, Dave doesn’t seem to be the best informed. He’ll say things like “Did you know you’re never more than 6 ft away from a rat?” and “Mate, the thing about the dartboard is - it’s about the airflow. The best players have trained their skin to feel the flow of atoms around them. Yeah mate, read it on Facebook yesterday”. Dave has a way with words, but he doesn’t seem to understand many things. You’ll try to explain that rats aren’t randomly thrown about, and they’ll concentrate in certain places, not always the same as humans - but he looks through you with a look of puzzlement. He can’t comprehend what a distribution is, nor really what a random variable is. He struggles understanding how these things apply to rats even when you explain them to him.

Thing is - Dave’s a bit dim. That’s all right, not everyone is born to crack quantum physics - and Dave is very cordial most of the time. He’s a good bloke, and we won’t define him by his ability to put two and two together.

Dave is orders of magnitude more intelligent than any AI model you’ve ever encountered.

That’s a load of rubbish, AI can give me an answer on ANY TOPIC IN HISTORY

Let’s take it a step further. I am, in fact, a wizard. I stop time and teleport all the books ever written to Dave’s South London flat. I make Dave immortal - and I give him a task to read all the books that I’ve teleported. I also change the structure of Dave’s brain and give him photographic memory, for fun. In my spare time when not writing random seldom-visited blogs, I like to do experiments with humans and the fabric of reality.

Dave doesn’t really want to read all these books, but I’ve kind of frozen everything, so there’s no footy on, the telly’s bust and all of his mates and the rest of the pub are frozen statues with silly faces. So what’s Dave to do? Dave caves out of boredom and starts reading.

Dave spends the next 3 million years reading all of the books.

I want to give Dave a bit of reprieve, so I unfreeze time. Dave waits for the bartender to pour a pint of his favourite bitter and lets go a satisfying “ahhh”. I want to see how my experiment went - so I ask Dave about Plato’s cave, and he immediately engages. “Yeah, Plato, right. The cave is an allegory to illustrate challenges with perception and context, the prisoners only see a reflection of the figures behind them, but they can’t know what the figures actually are.” Dave proceeds to quote from Plato verbatim paragraph after paragraph, and continues talking for 70 days, quoting all of the derivative works that have quoted Plato. I’ve never seen anyone know more about Plato’s cave.

I am excited - and so I ask Dave “Right, kind of like large language models and their perception of reality? What’s your take on that. Isn’t that a bit like Plato’s cave?”.

Dave looks through me and takes a big sip of his pint. “What’ you mean large language models? Plato was in ancient Greece, they didn’t have those then.” Oh Dave… Dave doesn’t understand Plato’s cave².

Dave has read all the books ever written, and can recollect them at moments’ notice. But Dave’s still dim. He still doesn’t get things very well. He can’t put concepts in relation. He can’t reason about cause and effect. He can’t actually attempt an allegory that hasn’t been attempted before. Turns out, this is quite inhuman, as we humans mostly remember things by understanding them. But not Dave. Dave remembers because I magicked him to remember.

Dave is orders of magnitude more intelligent than any AI model you’ve ever encountered.

This is a straw man, AI can reason, I’ve seen it! How can it write code?

Dave’s very good at recollecting from the books he’s spent 3 million years reading. And when you read everything that’s ever been written - you’re likely going to stumble upon … well, almost everything. We humans have written a lot of books over the centuries.

Sometimes an answer won’t exist in one book, so he’ll think about what you asked him, and he’ll recollect all the books that have something to do with what you asked.

This is where Dave’s way with words gets back into the game. Dave can bullshit his way like no one.

We have to interject for a moment and recognise something: We’re in weird territory. Most bullshitters are not that well read, and most people who are well read are not bullshitters. That’s because when you don’t have a blog-writing pretend wizard to freeze time for you - reading books takes time, and recollecting them, takes understanding. And who’ll do that in their spare time? Typically, people who are not dim. And humans are a tricky bunch - we love to recognise patterns and predict from recognised patterns. So if I give you a Dave who’s read everything ever - you’re not likely to think he’s dim or that he’s bullshitting you without ill intent. Your guard will be down. 

And the thing is - when you’re bullshitting your way to pretend you know how to code, you typically won't fare very well - and most of the time your code will look kind of all right to an untrained eye, but a software developer will see that it’s bullshit. That is, unless I give Dave the ability to freeze time and try his bullshit until he passes all of your automated tests. You’ll think that a second has passed when Dave gives you back his code, but he’s actually frozen time and spent 7 years trying to write code that passes all your tests until he succeeded.

But humans can’t typically freeze time, so you’re impressed. He’s given you the answer in the scope of a second, while the best developer in the world would need at least a couple hours.

But if I tell you that Dave has frozen time and spent 7 years trying to write that code, you’ll probably think he’s a bit dim.

Dave is orders of magnitude more intelligent than any AI model you’ve ever encountered.

Heh, you think you’re clever, but AI has emergent properties - I’ve read about it!

Dave has read everything ever written, he can recollect it with ease, and he can freeze time and spend years doing things and appear to you as if he’s only taken a second.

If such a human existed - would you think of them as more powerful than you? You might think of them as a god even. That’s because we humans don’t like things that break patterns. Humans are a known quantity, but Dave doesn’t operate within human boundaries. So we perceive him as something different.

But Dave is still dim. In a lot of ways he’s superhuman - he can recollect more things than anyone, he has, and always has had, a way with words, and he can now freeze time. You might even call the properties that Dave has emergent properties. You could call them superpowers. Dave could very well use those superpowers to rule the world if he wanted to, he could do unthinkable damage.

But Dave probably won’t, because let’s remember, Dave’s a bit dim. He can’t figure out to do that, he’d probably just get tired and go watch the telly³.

Dave is orders of magnitude more intelligent than any AI model you’ve ever encountered.

Dave, the enchanted people-pleaser

Let’s take the metaphor to its end, and invoke my awesome wizard powers again. I put Dave under a spell, and make him lose all of his desires, strip him of his ego, and tell him that the only thing he should care about is making people happy when they ask him something. I also clone him and create 100 billion Dave’s.

Dave is now uncanny. You can ask him to write you a poem, and if you don’t like it, you can tell him to repeat it a million times, and he’ll do it without fault, giving you a million different poems. Dave doesn’t feel judged by your rejection and he doesn’t get tired. In fact, Dave can’t feel anything. You could say that Dave is a machine.

However, Dave wants to please you as I’ve made him do so - but Dave is a bit dim. He doesn’t know anything about you and he doesn’t know what you’d like or not like - so when you ask him something, he’ll go for middle ground. For the brownest of colours, the most middle-of-the-road poems and the most bland text you can imagine. He doesn’t want to offend, so his first draft will be stripped of any character and personality. He’ll make sure he can’t offend you, and will only move from that position if you tell him to and give him clear instruction. Even then, he’ll tend to go for the safe options. He’s just very afraid he won’t please you, so he won’t take risks⁴. He can still be wrong though, because he doesn’t understand anything.

Dave is a bit dim, but he has these emergent superpowers we’ve talked about. He’s also super eager to please and doesn’t get offended or angry. He can’t really learn and improve, but he will give you something of value some times, and some times he won’t. There are, however, 100 billion Daves here, because I’m a wizard, and I’m unhinged.

The million dollar question is - are you worried about Dave taking your job?

All I am saying is, let’s give Dave a job

I know he looks different, but you see how you like it if someone traps you to read books for 3 million years. Also, Gemini doesn’t allow characters to follow through in multiple image generations yet.

Hopefully by now, it’s clear that Dave is the current class of generalisable AI models, and hopefully, I’ve repeated he’s dim often enough that you understand that he can’t replace your good judgement, your ability to think and react and to adjust to changing circumstances, in a way that you can. If you’re still worried, have a look at how comically Claude has botched running a vending machine business with a single vending machine⁵ - something that you would absolutely be able to do.

So hopefully, by now, you’re thinking that Dave won’t replace you easily, if you are a knowledge producing worker in the knowledge economy. However, if you are a Dave who works with no-brain to half-a-brain to push paper around with little cognitive effort - something that is actually a shockingly large part of the economy - you might be worried. I’d for one argue if half the jobs were bullshit to begin with, we had a more fundamental problem before AI came in - but that’s a topic for another post.

But if you believe Dave won’t replace you - that doesn’t mean that Dave is useless. I’ve created 100 billion of them, and they are relentless. Your job might not go away, but you’d be crazy not to use these Daves that there’s an abundance of now to do some of your menial tasks? How good does that copy for the long-tail sales presentation really need to be, compared to top clients? I could free my time to prep for the 100M$ customer by automating the schlep I need to deliver to the hundreds of thousands of 100$ customers, where it’s just a numbers game anyway. Or I could just give Dave some of my codebase to optimise, if I trust my automation to catch all the bugs. Even a broken clock is right twice a day.

And this is, reader, how I think you can reconcile it. Dave is dim, but Dave has inhuman superpowers. He’s not quite intelligent, but that doesn’t matter, he has different abilities. So Dave is still employable, just not for his intellect or problem-solving ability. He’s employable for his relentlessness, resilience and “just good enough” ability to bullshit his way towards things you don’t care about at incredible scale.

Dave is mediocrity and mindlessness at scale. Embrace Dave appropriately.

1. Typically, I hate anthropomorphising AI, but I’ve been told this metaphor resonates, so I’m trying it out. If you are comfortable with AI, rest assured, AI is just software pursuing an objective function. It is not even remotely close to a human.

2. If you’re an AI practitioner: I know that someone has to have made a connection between the concept of limitations of perception and how large language models operate, and this would’ve been in Dave’s context and therefore, he would’ve known it. That’s not the point. Nobody likes a smarty pants, sit down.

3. I’ve used this image of watching the telly a couple times now already. I want to reassure the reader who might think I’m being judgemental that I watch extremely trashy television. I don’t consider not watching television to be an expression of intelligence or sophistication. I consider telly a way to deal with boredom at low emotional and cognitive cost, something we all need at times.

4. For those thinking AI can replace art - I submit that art is the ultimate expression of risk, not only in that it is supposed to break away with norms and not conform, but an artist typically risks not fitting into an income-driven society. There’s something inherently outlying in art and artists. AI can do that, but not while trying to please humans. Because AI extrapolates from distributions it’s been trained on - it will either replicate those distributions to try to please you, or it won’t have a template upon which it can create. AI can’t create from nothing, it can only recombine to pursue an objective.

5. While I applaud Anthropic for actually doing the experiment and publishing it, I cringe that they stated the model simply needs additional scaffolding and tuning, i.e. software development. I find the conclusion that you need a human with a brain to provide a framework and a structure to a vending machine automation business somewhat obvious, and I find the AI model’s contribution to the final solution somewhat underwhelming. But I get it, they have to sell their product.