Gemini 2.0 Flash Thinking - Does it Pass the Misguided Attention Test?

[@elawchess]

Please can you be showing us what it does when you warn it that it's a trick question etc, whether it still gets stuck or not?

[@jamesjonnes]

From my tests even 4o is smarter than Gemini 2.0 Flash. OpenAI has fixed many simple mistakes that Google has not yet.

[@nubboi21]

No, actually im using the flash and it’s actually far far better that 4o , I use it for my science questions in physics and stuff and it’s really a lot lot better

[@elawchess]

Cos you can trick a human too with some of these and it wouldn't warrant the conclusion that the tricked human can't reason. If when alerted that it's a trick question it still can't do it then I'll probably agree about the seriousness of the issue. I've seen a couple videos when you've done this type of thing and seemingly concluded that "they can't reason", and I feel like that conclusion is not warranted.

[@Heisenberg2097]

As long as it never learns to think like you... humanity is save.

[@fabriziocasula]

wow

[@carlkenner4581]

I've met many humans who can't pass the misaligned attention test.

[@jeffwads]

Orion dropping tomorrow. Wait until you get a load of that model.

[@chamikk90]

it's capable, but not smart as o1

[@Cine95]

Its flash bro

[@samuelgarcia1802]

Yhea it's like the equivalent of o1 mini I suppose

[@NakedSageAstrology]

You can get even better results than 01, if you use an API and have it prompt itself back and forth.

[@josemarques3454]

yes!... and it's free.

[@ankitnmnaik229]

​@@NakedSageAstrology it's exterminatal and flash.. Not pro or ultra or specifically a separate reasoning model at all...

[@punk3900]

o1 pro is the 🔝

[@anubisai]

Okay moneybagg

[@deepbrar1]

so o1 did solve this problem ?

[@ankitnmnaik229]

200$