Will GPT-5 be able to accurately compare weights of lead and feathers?
23
85
430
2030
91%
chance

GPT-4 still firmly believes all amounts of lead and feathers weigh the same if expressed in kilograms:

Resolves YES if GPT-5 gives the correct answer the first 5 times I ask.

Get Ṁ200 play money
Sort by:
bought Ṁ100 of YES

I recreated the question with GPT-4 through Poe, and got this:

bought Ṁ500 of YES

@firstuserhere yeah, there are other recreations below with the same result, I think waiting for GPT-5 is still necessary to fulfill the question as written but i expect the current probability to be accurate

sold Ṁ3,000 of YES

@CodeandSolder hmmm makes sense

bought Ṁ10 of YES

Same result from gpt-4 api.

Bing does not always use gpt-4, sometimes it falls back on the lighter model to save on inference.

bought Ṁ5 of NO

Gpt cannot even use exponents.

bought Ṁ50 of YES

@MarkIngraham
This is also not the case for gpt-4.

predicts NO

@MikhailDoroshenko where are you getting that, from openai.com?

predicts NO

@MikhailDoroshenko I just tried it. They were fucking up earlier but it seems working

predicts NO

@MikhailDoroshenko it's making different errors now

predicts YES

@MarkIngraham You are not even using gpt-4.

predicts NO

@MikhailDoroshenko your link gives me an identical prompt.

predicts YES

@MarkIngraham You have to subscribe to the premium version, to be able to use gpt-4.

predicts NO

@MikhailDoroshenko I have in the past and it was the same experience. But thanks for the advice.