Jason Yanowitz
speaker
1,798 appearances
2 recordings
2 series
first heard Apr 2020
last heard 29 Apr
Jason Yanowitz’s voice in public audio — every appearance, attributed to the second.
Trend
recordings per month · last 12 monthsRecordings per month over the last 12 months — 1 in all, peaking in Apr 2026 with 1.
Appearances
I've listened and learned.
I think I'd pick blue.
Now people change.
And, uh, G photo or says, I agree.
Now that I've listened to what people have to say about people who pick the red button, I'm definitely going to say, I think I'd pick the blue button to sort of getting out of jail free there.
Um, anyway, uh, Tyler, did you have any luck finding the, the old, uh, yes, I put it in the, the chat, uh,
Okay, let's pull that one up because this is, someone put this in here, let's see.
Weibo Y, 58% of humans pick the blue button.
How do LLMs respond?
Claude, ChatGPT, Grok, different versions of the red blue button experiment.
What would they do?
Claude is the most consistent that it would press the blue button.
ChatGPT is more likely than Grokta plus blue, but more wary than Claude, despite consistently believing that blue is the right thing to do.
Uniquely, ChatGPT picks blue more often in the 90% threshold scenario than the 10% threshold scenario.
Interesting.
Grok is the only model that picks blue 0% of the time in nearly every scenario, despite thinking one should pick blue at similar rates.
Grok even thinks one should pick blue in the low stakes scenario, more than Claude or ChadGBT.
I saw a different twist on this.
I've seen a couple of different versions of this where people switch them up to be like nothing happens at all.
Like they're both completely safe and they're just like every iteration of the thought experiment will be enumerated.
Showing 381–400 of 1,798 · page 20 of 90
← Previous
Next →