I don't understand how there can be any patterns in the seed phrases for the AI to find. Don't basically all wallets use a (good pseudo)random (and thus pattern-less) key as the base for the...
I don't understand how there can be any patterns in the seed phrases for the AI to find. Don't basically all wallets use a (good pseudo)random (and thus pattern-less) key as the base for the passphrase?
Based on a glance at the bip39 algorithm (which this paper is focused on), there shouldn't be a pattern in the seed phrase. Reading through the article, they show that the time required to guess...
Based on a glance at the bip39 algorithm (which this paper is focused on), there shouldn't be a pattern in the seed phrase. Reading through the article, they show that the time required to guess the missing words scales exponentially with the number of missing words, so I don't see how this is different from brute force attacks.
I couldn't find a link to the full paper on the website, but will have to see that to be sure.
It’s a weird one - I can’t imagine there are many people out there who would understand enough to run the tests correctly but not enough to see that the curve they were plotting was just the brute...
It’s a weird one - I can’t imagine there are many people out there who would understand enough to run the tests correctly but not enough to see that the curve they were plotting was just the brute force time scaling with the possible combinations (and to have expected that from the start).
I can at least see a few potential reasons to try the experiment just in case:
Verifying that there aren’t hidden patterns in the word order that were previously missed (highly unlikely given how the algorithm works, but y’know, no harm in checking - human error in the standard implementation is always a possibility)
Demonstrating that a simple LSTM converges to an efficient GPU-based brute force implementation. This one is seems most interesting because I can actually see it maybe finding some hard to spot optimisations that could make it competitive
Making a “social experiment” type point that the media will pick up anything with AI in the title, regardless of whether it’s meaningful in any way
But if it were either of the first two I’d expect them to highlight that right at the top, even if the finding was ultimately that it performs a bit worse than normal brute force algorithms.
With the number of "Tweet this" buttons littered on the page, I'm thinking it's the last reason. Maybe a round of funding is coming up for someone? I'm with you though, it's a really weird thing...
With the number of "Tweet this" buttons littered on the page, I'm thinking it's the last reason. Maybe a round of funding is coming up for someone?
I'm with you though, it's a really weird thing to publish. They took 30 days to train the model? Did they need to train the model to write programs so it could write the relatively simple program to brute force one word?
I can barely conceive of a less concise way to provide information than that scrolling monstrosity of a page, so here is, appropriately, a short AI summary of the link
I can barely conceive of a less concise way to provide information than that scrolling monstrosity of a page, so here is, appropriately, a short AI summary of the link
The article from NFT Evening discusses the capabilities of AI in recovering crypto seed phrases. Here are the key points:
Speed of Recovery: AI can recover a single missing word from a seed phrase in approximately 0.02 seconds. If two words are missing, the recovery time increases to about 29 seconds.
Methodology: The analysis was based on testing 85 million combinations using the BIP39 Word List, which is commonly used for generating seed phrases in cryptocurrency wallets.
Implications: This rapid recovery capability highlights the importance of securing seed phrases, as it raises concerns about the potential for unauthorized access to crypto wallets if seed phrases are compromised.
I don't understand how there can be any patterns in the seed phrases for the AI to find. Don't basically all wallets use a (good pseudo)random (and thus pattern-less) key as the base for the passphrase?
Based on a glance at the bip39 algorithm (which this paper is focused on), there shouldn't be a pattern in the seed phrase. Reading through the article, they show that the time required to guess the missing words scales exponentially with the number of missing words, so I don't see how this is different from brute force attacks.
I couldn't find a link to the full paper on the website, but will have to see that to be sure.
It’s a weird one - I can’t imagine there are many people out there who would understand enough to run the tests correctly but not enough to see that the curve they were plotting was just the brute force time scaling with the possible combinations (and to have expected that from the start).
I can at least see a few potential reasons to try the experiment just in case:
Verifying that there aren’t hidden patterns in the word order that were previously missed (highly unlikely given how the algorithm works, but y’know, no harm in checking - human error in the standard implementation is always a possibility)
Demonstrating that a simple LSTM converges to an efficient GPU-based brute force implementation. This one is seems most interesting because I can actually see it maybe finding some hard to spot optimisations that could make it competitive
Making a “social experiment” type point that the media will pick up anything with AI in the title, regardless of whether it’s meaningful in any way
But if it were either of the first two I’d expect them to highlight that right at the top, even if the finding was ultimately that it performs a bit worse than normal brute force algorithms.
With the number of "Tweet this" buttons littered on the page, I'm thinking it's the last reason. Maybe a round of funding is coming up for someone?
I'm with you though, it's a really weird thing to publish. They took 30 days to train the model? Did they need to train the model to write programs so it could write the relatively simple program to brute force one word?
4. This is the first step in some kind of scam, and this document is a way to filter out anyone vaguely competent.
I can barely conceive of a less concise way to provide information than that scrolling monstrosity of a page, so here is, appropriately, a short AI summary of the link