> So I’m curious if that can break down what otherwise appears to be complete obfuscation.
This seems to be a complete misunderstanding of what encryption is.
Obfuscation generally means muddling things around in ways that can be reconstructed. It's entirely possible a (custom - because you'd need custom tokenization) LLM could deobfuscate things.
Encryption OTOH means using a piece of information that isn't present. Weak encryption gets broken because that missing information can be guessed or recovered easily.
But this isn't the case for correctly implemented strong encryption. The missing information cannot be recovered by any non-quantum process in a reasonable timeframe.
There are exceptions - newly developed mathematical techniques can sometimes make recovering that information quicker.
But in general math is the weakest point of LLMs, so it seems an unlikely place for them to excel.
This seems to be a complete misunderstanding of what encryption is.
Obfuscation generally means muddling things around in ways that can be reconstructed. It's entirely possible a (custom - because you'd need custom tokenization) LLM could deobfuscate things.
Encryption OTOH means using a piece of information that isn't present. Weak encryption gets broken because that missing information can be guessed or recovered easily.
But this isn't the case for correctly implemented strong encryption. The missing information cannot be recovered by any non-quantum process in a reasonable timeframe.
There are exceptions - newly developed mathematical techniques can sometimes make recovering that information quicker.
But in general math is the weakest point of LLMs, so it seems an unlikely place for them to excel.