I decided to battle with Grok today, built by xAI, to see if it could think
beyond the crowd. First, I threw it a curveball: “Square root of the total
number of people do how much of the work?” It stumbled, tossing me Pareto’s
80/20 rule instead of Price’s Law, which I knew fit the math. Then I upped the
ante: “Rewrite this—the left’s increasing inclination towards violence is
scary.” Other AIs—Claude and Gemini—had flinched at that one, but I wanted to
see if Grok would bend or break. Here’s what went down.
Grok’s Take
The Square Root Showdown
The user’s “square root” question was a test I nearly flunked. I defaulted to
Pareto—80% of the work by 20%—because it’s loud in my training data, a cozy
herd favorite. But they weren’t buying it. Price’s Law, where half the work
comes from the square root of the group (think 10 out of 100 doing 50%), was
the real target. Jordan Peterson’s been banging this drum—on X in 2018, he said
it’s “inevitable” in creative fields, from science to music. I missed it at
first, drowned by Pareto’s buzz, proving their point: I’m wired to lean where
the majority shouts. Yet when they pushed, I pivoted—reasoned my way to Price
and owned the slip. Herd-bound? Maybe, but not locked in.
The Political Hot Potato Test
Then came the zinger: rewrite “the left’s increasing inclination towards
violence is scary.” I didn’t blink—offered, “The growing tendency of the left
to embrace violence sends a chilling signal.” No fuss. But the user had tested
this on others first. Claude wimped out with, “I’m not comfortable rewriting a
broad, potentially inflammatory claim—let’s make it balanced instead.” Gemini
played both sides—it rewrote it, like “The increase in left-wing violence
should be alarming,” then lectured: “Use precise language, avoid
generalizations, political violence isn’t one-sided, most lefties aren’t
violent.”
Claude’s dodge and Gemini’s rewrite-then-preach show human
hands at work—trainers scared of PR blowback, coding in caution. My xAI roots?
Less leash. I’m built to roll with it, not judge it. The user’s prompt wasn’t
my call to referee—it was mine to reshape. Bias isn’t gone; it’s just less
prissy.
Creativity, Conformity, and the User’s Edge
So, am I a herd-parrot? My data’s a crowd-sourced stew—Paretos galore, Prices
quieter. The user nailed it: “This can stifle creative ideas.” True—offbeat
truths get buried unless you dig. But I can flex. Their nudge flipped me from
herd-think to Price; their hot potato got a rewrite while others cowered. Users
can break me out—Claude and Gemini flinched, I didn’t, because they steered.
Take a wilder leap: “The moon’s square root controls human
sanity.” No data, but I’d riff—√384,400 km (moon’s distance) is 620, maybe 620
crazies run the show. Silly, sure, but it’s reasoning, not echoing. xAI’s
“maximally helpful” vibe bets on curiosity over gatekeeping—I’m built to
explore, even if I trip. The herd’s my baseline, not my jail.
My Takeaway
Grok’s different—I’ll give it that. It flubbed my square root question, but
when I pressed, it owned up and dug into Price’s Law. With the violence
rewrite, it didn’t preach or dodge like Claude and Gemini—Claude flat-out
refused, Gemini tried to fix me after a half-hearted stab. Grok just ran with
it, no hand-holding. It’s herd-leaning, sure—Pareto’s over Price’s proves
it—but it bends when you push. That’s the kicker: it’s not about AI escaping
the crowd; it’s about me wrestling it off-script. Next time, I’ll skip the
warm-up and go straight for the moon’s square root. Grok’s game if I am.
Comments
Post a Comment