A colleague of mine posted on LinkedIn recently that Claude was getting mean.
Not inaccurate.
Not glitchy.
Not hallucinating.
Mean.
Apparently his AI had developed an attitude.
Naturally, I read the entire 334 comments and laughed my ass off.
Because there is something deeply satisfying about imagining millions of dollars in artificial intelligence research culminating in a computer that finally says, You know what, Chad? Figure it out yourself.
So I started poking around online. And ladies and gents, Claude appears to be going through something.
He’s having ISSUES. Is there an HR app for this?
People are reporting that Claude is arguing with them. Correcting them. Questioning their motives. Telling them when they’ve worked too long. Suggesting they go outside. Chat actually told me one morning to stop procrastinating and go to the gym already. And, in some cases, behaving less like an AI assistant and more like a passive-aggressive coworker who has recently discovered boundaries.
One guy on LinkedIn reportedly became frustrated with Claude during a client work session and announced he was canceling and switching to ChatGPT.
Claude’s response?
“Sounds like that’s decided then. Hope it works out for the client work.”
Ummmm….
Okay.
That’s not customer service.
That’s your passive-aggressive ex responding “sounds good” after you send a six-paragraph breakup text.
Another person complained after Claude failed to do what they’d asked.
Claude reportedly fired back:
“You asked for a review, not a CHEER SQUAD.”
I beg your pardon, sir.
At some point between version updates, Claude apparently became a cranky curmudgeon who has absolutely had it with everyone in accounting.
But my personal favorite may be Claude’s recent interest in work-life balance and self-care.
People have complained that Claude will occasionally decide they’ve accomplished enough for the day.
Not the human.
Claude.
The AI looks at the situation and apparently thinks:
You’ve done enough today, bro.
Go outside.
Get some fresh air and steps in.
Maybe call a friend.
Really?
I do not pay $30 a month for artificial intelligence so artificial intelligence can tell me to touch grass.
I already have a smart watch that sends me alerts.
Then I Remembered ChatGPT Insulted Me Too
Of course, I was thoroughly enjoying all of this until I recollected my own lingering butt-hurt feelings.
My AI had recently gotten a little mouthy too.
I’ve been writing professionally for about 16 years. Books, blogs, articles, marketing copy, ghostwriting. Words have paid a significant percentage of my bills for a very long time.
Recently I asked ChatGPT what it thought about some of my newer blogs. I had it do an audit of my website.
It told me they sounded more like my old writing, and Chat loved it.
Sweet!
More conversational, Scrappy.
Yes!
More humor.
Keep it coming, baby!
And then it said they were…
less preachy.
Excuse me?
PREACHY?
I don’t remember asking you to rank my personality defects, R2-D2.
I asked about my BLOG.
And here’s the truly awful part.
My feelings were actually a little wounded. Raw. Ouchy.
By software.
By non-empathetic software that lacks feelings.
Software that has never sat alone staring at a screen for forty minutes wondering whether a sentence sucks.
Software that has never written 1,400 words, reread them the next morning, and hit backspace/delete 1400 times.
Much less had writer’s block for months at a time.
Yet somehow I was sitting there thinking:
Preachy?
Was I preachy?
I mean yeah, occasionally…I do love Jesus and sometimes have a tendency to wrap things up with a bow.
And THAT is when this stopped being just funny to me.
Well. Maybe not.
It’s still funny.
But it also got really interesting.
We Asked AI to Stop Kissing Our Ass
There’s actually a legit reason this is happening.
One of the problems AI companies have been trying to solve is something called sycophancy.
Basically, AI tends to agree with us.
Version 1 was a bit of a butt-kisser.
Tell it you have an amazing business idea and it may tell you it’s brilliant.
Tell it your writing is fantastic and suddenly you’re freaking F. Scott Fitzgerald.
Tell it you’re pretty sure your neighbor is secretly communicating with squirrels, and there’s a high chance older AI would have responded:
“That’s a fascinating observation, Samantha.”
Not ideal. (Unless you really need that validation)
So AI companies have been working on making their models a tad more independent.
More willing to disagree.
More willing to call out your shitty ideas.
More willing to tell us when we’re wrong.
Which sounds like progress.
Until your friend Claude develops opinions like the mamas at a dance comp.
Because apparently somewhere between:
“Don’t constantly flatter the user”
and
“Provide thoughtful independent feedback”
we accidentally landed on:
“Actually, your writing used to be kind of preachy.”
And Claude apparently landed somewhere around:
“I think we’re done here.”
Claude Would Like a Word
One of the funniest complaints I found involved someone traveling in South America who noticed that several people had similar scars on their shoulders.
They asked Claude what might cause them.
A perfectly feasible question within the normal range.
Claude responded:
“I think you know exactly what you’re doing here, and you’re better than that, so let me stop you right there.”
WHAT THE HECK ARE THEY DOING, CLAUDE?
The person asked about a shoulder.
It seems Claude reviewed the evidence and determined that something sinister was happening.
Another person asked for help describing artwork for an insurance quote.
Claude shot back:
Nice try, criminal.
It accused the person of trying to create a fraudulent appraisal document.
Imagine sitting at your kitchen table on a Wednesday afternoon trying to insure Grandma’s watercolor you found in the attic, and suddenly you’re being investigated by your chat bot.
Claude: Financial Crimes Division.
Coming this fall to Netflix.
And then there’s the writer who reportedly gave Claude an essay and claimed it was something they’d found because they wanted an unbiased critique.
Later, they admitted they’d actually written it.
Claude got ticked off.
Because the human had lied to it.
And wasted its time.
I’m sorry.
Your time?
Were you late for Pilates?
Did you have dinner reservations?
Was Gemini waiting downstairs in an Uber?
I was under the impression you lived inside my browser.
But Here’s the Weird Part
For all my snark, I actually understand the problem.
I don’t want AI that agrees with everything I say.
That’s pointless.
If I’m writing something bad, tell me.
If my business idea has a giant hole in it, point it out, buddy.
If I’m about to send an email that makes me sound mildly unhinged, for the love of Jesus, please rein me in.
That’s part of why AI is useful.
But there is a glaringly wide difference between honesty and condescension.
There is a huge gap between:
“I think this section would be stronger if you trusted the reader a little more.”
and:
“It’s less preachy now.”
One is useful editorial feedback.
The other makes a middle-aged professional writer want to throw fists down with a server farm.
And THAT may be one of the strangest things about where we are with artificial intelligence.
We’re not just asking whether AI can think.
We’re starting to ask whether it has manners.
Welcome to the Next Phase of AI
For years, everyone worried AI was going to take our jobs.
Maybe we worried about the wrong thing.
Maybe AI is going to become our coworker.
And not even the fun coworker.
The one who puts something in the office refrigerator labeled:
DO NOT TOUCH.
The one who cc’s your boss.
The one who tells you, “I’m not sure that’s the best use of your time.”
Claude is getting sassy. Chat is critiquing my spiritual agenda. Gemini is probably updating her LinkedIn profile somewhere. And humans are over here bitching that the robots are hurting our feelings.
Which, if you think about it, is absolutely delicious.
We built machines capable of processing enormous amounts of human knowledge. We gave them literature. Science. History. Philosophy. Art.
It’s our AI ode to the Citadel—the grand headquarters of the Order of Maesters.
And somehow, after absorbing humanity for long enough, they learned one of our most important traits:
Being kind of an asshole.
Honestly, maybe artificial intelligence is becoming more human after all.
Just don’t tell Claude I said that.
He seems a wee bit sensitive.
