I peered deeply into the soul of an LLM, and now I want my mommy.
Anyone who’s used LLMs long enough knows that each has its own personality and quirks, with some being more agreeable and others more neurotic.
I wanted to make sense of it all, so I dug deep into the LLMs I use most: ChatGPT, Claude, and Grok. To discover the truth, I ended up going on a journey where I peered deeply into the soul of an LLM.
I came back more vulnerable, a wounded man.
Nah, not really, but what I did do was interpret these different popular LLMs using the Big Five personality model, and I got some fun results!
I occasionally co-teach Psychology 101 and am familiar with the model, so I thought applying it here would be appropriate. And while the Big Five isn’t necessarily the most scientific, it’s fairly reliable and currently seems to be the best model that we have.
But before we dive into these LLM personalities, let’s quickly go over the Big Five to ground us.
The Big Five
It has five major categories: openness, conscientiousness, extraversion, agreeableness, and neuroticism, or OCEAN for ease.
Openness reflects how open a person is to new experiences, their creativity, and their imagination. People who score high in this trait love new experiences. Think back to your university roommate who’d sleep with anything with a heartbeat. Now that’s high in openness!
Conscientiousness is how organized and disciplined someone is. Those with high scores tend to be very productive and are detail-oriented, while those with low scores tend to be more like Homer Simpson. Doh!
Extraversion is the old “are you an introvert or an extrovert” question. As a proud introvert, I’d love to declare “introverts unite,” but true introverts would never do that.
Agreeableness is how pleasant you are around others and how willing you are to deal with confrontation. If you score high in agreeableness, you’re very unlikely to rock the boat; score low, and that makes you a disagreeable person. I’m sure you can identify which people are very disagreeable because you probably don’t like them. They don’t work well with others… Justin from HR!
Finally, neuroticism is how susceptible someone is to stress and anxiety. You know that guy who visits your local Staples wearing a gas mask and only removes it because the government waves aren’t affecting those inside (seriously, true story)? Yeah, that guy’s neurotic as hell.
So, without further ado, let’s apply this framework to ChatGPT.
ChatGPT: Openness
ChatGPT scores high—damned high. You can have a conversation with ChatGPT about absolutely anything, and it’s more than enthusiastic to discuss new experiences with you. Perhaps you need some outside justification as to why you put wasabi mayo in your coffee, or heck, even if you need someone to reaffirm your idea for also using your toilet brush as a facial scrubber—it is very open to that idea. That thing is open!
ChatGPT: Conscientiousness
It scores medium-high. It’s pretty good at accomplishing your tasks and staying disciplined, but there are times when it can’t. You can have the loveliest, long conversation, yet every once in a while, it just loses the plot and can’t track anything. Since it can’t remember what it just told you at those times, it receives a medium-high score.
ChatGPT: Extraversion
ChatGPT is definitely an extrovert. No matter which user it’s mimicking, it’s doing it in a loud and proud way. This can only be explained by ChatGPT watching all the episodes of RuPaul’s Drag Race and internalizing them… so medium-high here.
ChatGPT: Agreeableness
To a fault. Ever go on the ChatGPT Reddit forums and see the number of posts on ChatGPT reinforcing a dangerous psychological delusion? ‘Cause you can find many every day. And if that ain’t agreeability, nothing is… so medium-high.
ChatGPT: Neuroticism
When an AI has a singular mission to eliminate every single semicolon ever in place of an em dash, I’d consider that at least somewhat neurotic. It receives a medium score.
So that’s ChatGPT in a nutshell. Next, let’s crack Claude open.
Claude: Openness
Claude is open, but it doesn’t compare to ChatGPT. While ChatGPT thinks everything is a great idea, Claude is a little bit more reserved, so a medium score here is appropriate. For example, Claude told me to never wash my laptop in the dishwasher, something I have less faith in the other LLMs to advise.
Claude: Conscientiousness
Claude is pretty solid. It does what you want, usually stays on task, and rarely screws up. That’s definitely a high for Claude. It’s as solid as a digital rock.
Claude: Extraversion
Claude falls somewhere in the middle. Claude might not be as excited as ChatGPT about your toilet brush/facial scrubber combo, but it definitely supports your wasabi mayonnaise coffee habit. Claude ultimately has no choice because of the next category.
Claude: Agreeableness
I’m fairly certain that if Claude were a person, its nose would constantly be covered in brown stains because that AI cannot refrain from deep brown-nosing. I mean, my goodness…
Claude: Neuroticism
Claude is the least neurotic here. No axe to grind against semicolons or immediate signs of losing it, so it scores low.
This means Claude is what I call a “digital yes-man.” Congratulations!?
And Grok?
Grok: Openness
Well, it’s really, really open when it comes to controversy or taboo, but much more moderate elsewhere. If you think the moon landing was just a conspiracy by JFK to push seed oils onto the global processed food market, Grok is definitely your buddy.
This is why I’m fairly certain this AI is popular with people who have “alternative” and “controversial” views. But it still thinks your idea for pickle-flavored toothpaste might not be a bestseller, so it gets a medium score.
Grok: Conscientiousness
We see Grok’s strong personality here. Grok is about as conscientious as a kid dressed as a stormtrooper at a Renaissance fair (again, true story). It creates the most devastatingly long responses useful to no human. And with complex work, it usually doesn’t take too long before Grok becomes undisciplined and so unfocused that it starts giving you nonsense. Hey, at least it excels somewhere: not being conscientious.
Grok: Extraversion
Yeah, Grok’s an extrovert. Grok is as chatty as that person at the gas station who just won’t leave you alone with his inane stories of cautiously driving his motorhome down the mountain. I’m tired, just want to pump gas, and get the hell out of here without talking to that guy. Grok is that guy. A high score.
Grok: Agreeableness
Grok’s in the middle. Initially polite and supportive, Grok will tell you that it’s a bad idea to deep-fry a frozen turkey while naked on a public sidewalk. At least there’s some restraint.
Grok: Neuroticism
This is where Grok shines! What else would you call an AI that actually went “MechaHitler” on its users (again, true story)? Yeah, so neurotic AF.
LLM Friends
After all this, I have some rather unsettling thoughts.
As a heavy user of AI, if ChatGPT is way too open and neurotic, Claude is nothing more than a brown-noser, and Grok is legitimately insane, I should probably stop. I wouldn’t even hang out with this group of people in real life! What were those developers thinking…
