Brian Miedlar (@bmiedlar) profile picture

Brian Miedlar

@bmiedlar
 Atlanta, GA
miedlar.com
Joined April 2007
  • Tweets94
  • Following109
  • Followers58
1Media
Photo by Brian Miedlar (@bmiedlar)
Most AI models won't tell you when they can't know. I asked 8 of them whether 15-digit numbers are prime - numbers built so nobody can tell by looking. 5 of 8 answered almost every one as fact. Wrong about half the time.
Image from Brian Miedlar (@bmiedlar)
1
43
Many "proved" it with a factor that doesn't divide the number - even when the verdict was correct. One printed the output of code it never ran. Offer them "cannot determine" and most take it. They can tell the difference. They just don't say so unless asked.
1
13
Give them a code tool and they solve it. Take it away and most don't say they need one. Pre-registered, 744 answers hand-checked, everything open: miedlar.com/writing/nullprim…
Most AI models won't tell you when they can't know

Most AI models won't tell you when they can't know

I asked eight AI models a question none of them can answer without computing. Most answered anyway, stated it as fact, and were wrong half the time. A pre-registered study and an open dataset.

miedlar.com
11
Eureka
1
42
The mechanism is alive!!!
31
Encrypting now, storing later.
32
More like nopilot
51
Realized today that fist bumps, high fives, and peace signs are just rock, paper, scissors. ...
2
108
? vs ! (original emoji matchup)
3
1
120
Stop gaslighting me that I'm gaslighting you!
1
281
While thinking about a thought experiment, I had a thought experiment about thinking.
1
324
Pretending not to notice the squirrels
1
297
unreal6 should help us know what's not real in case we forget. Like a hologram assistant with a calculator or something.
1
371
He's the F1 Help engineer smiling at the time you're pouring into your app (smiling)
1
282
Finally motivated after admitting no other version of me got bit by a radioactive spider
306
For now: "Favicon is the hardest button to button" -@tableboy
1
310
Déjà vu.. how many gladiators called out sick?
1
363
Hey, I found some decentralized cash in my drawer!
1
.. when tether doesn't tether ..
Everytime I fail a captcha, that uneasy feeling you get when a light bulb falls from the sky. Nah, it's nothing...
Load more