Can OpenAI's o1-mini Decipher my "Secret Messages"?

I test o1-mini's ability to break relatively easy ciphers that I "cleverly" came up with at 7 AM. Is it impressive? Maybe only a little bit right now, but in the future, is this a problem we need to seriously consider? Watch and make your own judgement!
Timestamps:
1. 0:26 First Cipher (“Hi I am Kyle”) with replacement of letters to number order in alphabet.
2. 1:35 o1-mini starts to decipher
3. 2:30 Second Cipher (Cesar Cipher)
4. 3:04 o1-mini attempts to decipher second message
5. 3:54 Providing a hint
6. 4:37 Final message
7. 5:58 o1-mini’s final attempt (no hints)
8. 6:55 Another hint
9. 8:19 o1-preview attempt (no hints)
10. 9:22 Post-video discussion on AI safety, codebreaking, and quantum computers

[@mrpro7737]

I think AI is incredible, but it still needs that human touch to really shine, especially when it comes to navigating anything that requires strategy. Whether it’s making complex decisions or adapting to unique situations, AI benefits a lot from human insights and guidance. When we pair our experience with AI’s capabilities, we’re able to create solutions that are smarter, more effective, and really tailored to the challenges at hand.

[@user-ze9tj9yj4t]

Are you an AI bot sir?

[@KyleKabasares_PhD]

@@mrpro7737 I agree! Humans’ breadth of skills and tasks combined with AI’s computing power is quite the duo.

[@mrpro7737]

@@user-ze9tj9yj4t I am not a Robot 😃

[@privateprofile3517]

ther is no thing as human touch, any human quality which can be quantified , AI will do it better very soon !

[@thenicesven5328]

"I thought AlphaGo was based on probability calculation and it was merely a machine. But when I saw this move, I changed my mind. Surely AlphaGo is creative. That move was really creative and beautiful." - Lee Sedol after seeing AlphaGo's famous move 37

[@xinpingdonohoe3978]

You know, the weirdest bit was how it extracted "H I I A M K Y L E" but outputted "HI I'M KYLE". It just decided to apostrophise your message.

[@bokuboke482]

Neat upload. Think this is why we'll need properly-aligned AI to protect us from malicious (or merely curious) AI that could crack open digital safes and disseminate their classified contents to any and all bad actors. I'm imagining ASI doing the safecracking. Cheers & HH!!

[@Fenixtremo]

Next videos: Kyle - “I was able to create a dangerous virus using o1” Scammers - “Great, you helped us out buddy”

[@KyleKabasares_PhD]

@@Fenixtremo LOL no!!! I would never

[@parthasarathyvenkatadri]

Every new version from now on is going to get the previous versions discussions to find it without clue

[@Mr__Potato_]

yeah, i think we're safe from AI for a while longer than everyone makes out

[@spadaacca]

Why would you use o1 mini and not o1 preview? Seems a bit meaningless to test the o-series abilities with the second-best model?

[@KyleKabasares_PhD]

Just didn't want to waste my prompts since I wanted to use o1-preview for something else that day. Plus I did try o1-preview on some of these messages in the video, and it didn't get it

[@amahara]

Do the zodiac killer cyphers with hints

[@christiancarter255]

Oh, that is a good idea. I didn't think about that.

[@HelloCorbra]

Great ideas, let me try this😂

[@marwin4348]

it will be insane when one day a new model update nonchalantly solves these decades old ciphers

[@Hshjshshjsj72727]

I tried this with old copilot and multiple languages in same line and all cyrillic and yes ai can do it quite well

[@abhrodipsingharoy4508]

Hey can you test another ai? Its called hanooman everest 1.0 ai. Its created in india. It doesn't have any benchmark but it seems good i mean it got snake and flappy birds game right in python on first try.

[@alexjos79]

This prompt works. What does below mean 26, 11, 2, 12, 22, 23, 11, 8, 22, 14, 2, 5, 15, 24, 8

[@alexjos79]

In o1 mini and o1 preview

[@KyleKabasares_PhD]

It didn’t return a deciphered message for me!

[@babydragoon]

Sky is the blue ?

[@AbelShields]

I feel both hints were pretty egrigious, but the first one in particular gave it away completely. Why not just tell the model it has to crack the code? This video seems pretty pointless. By the way, I'm fairly sure Claude can read base64 natively, I'm pretty sure it wouldn't struggle with the first code and I think it might get the second one too.