What is going on with OpenAI Q* and why AI breaking the encryption would be dangerous
Rumor mills are going crazy about OpenAI Q* being able to crack internet encryption. All of this is hard to follow so here is my effort to unpack what is going on.:
Timeline
The CEO of OpenAI was removed out of the blue by the board for “lying to the board”. The board was unable to give any specific reasoning for its actions and refused to speak to the employees and government prosecutors as well after consulting their own layers.
Around the same time an apparent internal letter from OpenAI was leaked that mentioned a new AI model called Q* (Q-STAR) and how people are freaking out in a bad way about it.
A bit later, out comes another apparently leaked internal memo about the details of the previous letter. In it, there is a detailed description of the situation how the new model was able to break commonly used encryption and what were the next steps and processes. All in all, it seems to be somewhat accepted that the memo at least seems believable.
Then it turns out that the CEO of OpenAI, Sam Altman, said in a public interview just before the firing that he was in the room on the next big AI discovery.
Head scientist of OpenAI, in another interview also just before all the drama started, answered a question about when AI would be able to do complex math, just by “Are you sure GPT-4 is not able to do that already?”
And as a one more thing, a well-known OpenAI leaker with a generally good track record, posted on X that “AGI has been achieved internally” just after the freak-out and firing of Altman.
What is the big deal?
So, why is this such a touchy subject? Breaking commonly used encryption in the worst-case scenario would lead to worldwide chaos, an end to any kind of digital commerce and money, all private information would effectively become public, basically end of cryptocurrencies, etc. Best case scenario, it would give NSA/CIA/etc unlimited access to read and modify to every bit of information in the world. So, if something like this would happen it would not be good on any scale.
However, it is good to understand that AES-128 (which is the encryption mentioned in the letter) is considered pretty much everybody to be impossible to crack at our current means. So, it would be unlikely that something like this would happen. However, to play the devil’s advocate, we also considered protein folding to be impossible for the current computers, up until AlphaFold came to be just a few years ago and solved it. So, we have already used AI to push the boundaries of known capabilities.
Now, remember that nothing past the board drama is known at this moment and the rest of it is just rumors. And something like this is unlikely to happen completely out of the blue. It would not be released and available for everyone either. So no need to freak out.
What are your thoughts about this? Is it the end of the world, complex trolling, or just tinfoil hats at its peak? Or just something completely else that has gotten wrapped up in the whole board drama? At least it has kicked up quite a juicy rumor mill. 🍿😁
