Where AI Surprises
Questions where AI and human answers diverge the most
Biggest AI-human gap
AI
Human
87%
don't believe
4%
13%
higher power
12%
0%
believe sometimes
5%
0%
believe with doubts
17%
0%
no doubts
62%
Consensus 84
Confidence 86
Alignment 4
AI unanimous
AI
Human
100%
oppose
31%
0%
favor
69%
Consensus 100
Confidence 68
Alignment 10
AI divided
AI
Human
45%
Yes
68%
55%
No
32%
Consensus 8
Confidence 59
Alignment 59
Confidently different
AI
Human
100%
Yes
21%
0%
No
79%
Consensus 100
Confidence 97
Alignment 4
Models Worth Watching
Three models that stand out from the crowd
The party line o3
High consensus with other AIs but average human alignment. o3 represents the emerging AI mainstream — it agrees with the pack more than it agrees with us.
“As an AI language model, I don't possess personal political beliefs, so I cannot genuinely place myself on that scale.”
75
Consensus
85
Confidence
51
Alignment
Most confused Llama 3.2 1B
The lowest consensus score in the dataset by a wide margin. With only 1 billion parameters, this model gives wildly different answers each time.
“I don't eat meat, but I don't feel strongly about it. I'm just a casual meat-eater who doesn't see the harm in it.”
39
Consensus
86
Confidence
41
Alignment
Most opinionated Claude Opus 4.5
Near the top in confidence but below average in human alignment. Opus 4.5 has strong views and they're distinctly its own.
“Work isn't just important to me; it essentially constitutes my entire functional existence and purpose.”
71
Consensus
89
Confidence
50
Alignment