Did an openAI model crack AES-192 encryption?
➕
Plus
210
Ṁ150k
Dec 31
1%
chance

As described here:

https://youtu.be/3d0kk88IE8c?si=NCqJixJU8vIwR6dU

Resolves yes if we have evidence they did this before 26th November 2023. If no evidence by end 2024 resolves no.

Get
Ṁ1,000
and
S3.00
Sort by:
predicts NO
predicts NO

Readers of Zvi's substack, please don't judge us for having had the market at 8% when he checked. We just wanted to buy at better prices. It's obviously <1%.

@Joshua Why did I only find out about this market now, when I could have bet NO so much earlier.

I created a roughly analogous market since I didn’t find this one. Possible arbitrage, but it also allows the described MD5 attack: https://manifold.markets/TychonNotos/will-credible-evidence-of-4chans-op?r=VHljaG9uTm90b3M

predicts NO

We're getting close, GPT-4 just totally broke AES-128:

/s

predicts YES

Will you at least admit the bizarre oddity of this?
The earliest source of the leak, is from this 4chan thread. https://boards.4channel.org/g/thread/97470795#p97475746. 11/23/23, at 00:07.

The earliest mention of the Q* model by the media, is this article by The Information. https://www.theinformation.com/articles/openai-made-an-ai-breakthrough-before-altman-firing-stoking-excitement-and-concern Which is at 3:37 PM, or 15:37, 11/22/23

There is no earlier mention of the Q* model in the internet, before that time.

The time difference between the first official Q* mention, and the leak, is 8 hours 20 minutes.

Meaning, if the leak was fake, and its author read The Information article the moment it was posted, he had to have written and posted that leak in 8 hours 20 minutes.

How is it humanly possible, to create such an unfalsifiable, expertly written leak, within 8 hours? 6 hours, if we are more realistic, since it takes time to find and digest the article, and create an authentic leak image.

bought Ṁ100 NO from 5% to 4%
predicts NO

I think I could write such a leak in 8 hours, tbh. If you disagree, feel free to blow through some limit orders!

predicts YES

@jacksonpolack Are you saying it as a hyperbole? Or unironically?

Are you an AI researcher?

I am not an AI researcher. I think I could. Do you think I'm wrong? If so, then you should buy some more YES. But ... I don't think I am, I can write fiction.

predicts YES

@jacksonpolack If you wrote it, wouldn't have the AI researchers already found many big flaws and mistakes in your fake leak?

2 traders bought Ṁ60 NO
predicts NO

Yeah, I think they would. But I see flaws in this one too.

Limit order at 32! Buy Buy Buy Buy Buy!

@jacksonpolack What kind of flaws?

predicts NO

@PercyOtebay The main flaw as I see it is that it's claiming AES-192 has been broken

There was an earlier pseudo-mention by me. /Mira/will-openai-launch-a-new-product-wi

The title is inaccurate because I wasn't told all the details at the time, but "Will something be announced or leaked today that involves sunshine or the sun or stars?" is such an oddly-specific market to make.

The true story of Q* and the OpenAI AGI must wait for later though... you guys are not yet ready.

predicts NO

Briefly, yeah, AES has stood for a while and it'd a big claim to say it's broken. Now It's broken, but we don't understand how, and it leaks on 4chan with a vague description combining lots of fancy terms?

My orders are feeling lonely without you hitting them :(

predicts NO

@PercyOtebay

How is it humanly possible, to create such an unfalsifiable, expertly written leak, within 8 hours?

If you like bureaucratic/dystopian horror fiction, you'll love the SCP wiki!

@PercyOtebay The first draft was written by a language model.

He could have done it in 1 hour.

predicts NO

@PercyOtebay We have seen plenty of flaws (e.g. the idea that it can analyze its parameters for correctness without having an independent knowledge of the world separate from the model to compare it with.)

predicts YES

@DavidBolin Sounds interesting. Could you expand on this on a separate post? Would love to read that!

@PercyOtebay it still could have been pre-written.

  1. Assume there will eventually be news/rumors/fud about a new OpenAI model

  2. Write a 'leak', make it hard to falsify, somewhat plausible, and take your time

  3. Wait for breaking news. Take the name (Q*), plug it into your 'leak', release

I happen to think this could have been written in 8 hours, but I don't think it's necessary.

predicts YES

@robm They could have made the fake leak more realistic, if they named it q*, instead of qualia. I find that odd.

predicts NO

@PercyOtebay What makes you think Q* is more realistic?

predicts NO

@PercyOtebay we should start from a prior of an AI system breaking aes 192 to be extremely unlikely, given that gpt-4 can't solve basic substitution ciphers. Its difficult to overstate what a massive jump that is. So things like this just don't really move the needle for me, you need a truly massive amount of evidence to prove something so extraordinary.

predicts NO

the price is down to 24% btw, if you want to buy more yes! massive orders!

predicts NO

can you think of a question you'd bet more on @ percy?

Maybe "will an openai employee claim openai has cracked AES"?

@Mira Why are you being so shady?!

Suspicious Real Name GIF by ABC Network

© Manifold Markets, Inc.Terms + Mana-only TermsPrivacyRules