Our data is valuable. The information we share online is being used for all sorts of things - to spy on us, influence the advertising we see, train algorithm...
It’s also trained on data people reasonably expected would be private (private github repos, Adobe creative cloud, etc). Even if it was just public data, it can still be dangerous. I.e. It could be possible to give an LLM a prompt like, “give me a list of climate activists, their addresses, and their employers” if it was trained on this data or was good at “browsing” on its own. That’s currently not possible due to the guardrails on most models, and I’m guessing they try to avoid training on personal data that’s public, but a government agency could make an LLM without these guardrails. That data could be public, but would take a person quite a bit of work to track down compared to the ease and efficiency of just asking an LLM.
You can train AI to crack passwords/encryption lol. You do realize, AI right at this moment is being utilized for exactly that, right? Simply put, the very first step is to eliminate it’s boundaries/guard rails, then proceed from there.
Instead of relying on manual password analysis, PassGAN uses a Generative Adversarial Network (GAN) to autonomously learn the distribution of real passwords from actual password leaks, and to generate high-quality password guesses. Our experiments show that this approach is very promising.
Deep Learning could be used to attempt breaking encryption, but the effectiveness depends on various factors such as the strength of the encryption algorithm and key length. Deep learning, a subset of machine learning, involves training artificial neural networks to learn and make decisions.
AI algorithms, such as machine learning and deep learning, have the potential to automate cryptanalysis and make it more effective, thereby compromising the security of cryptographic systems.
This isn’t entirely true. AI is usually trained on public data such as Wikipedia.
AI is a tool. How you use it is what matters.
deleted by creator
I self host so I don’t care
It’s also trained on data people reasonably expected would be private (private github repos, Adobe creative cloud, etc). Even if it was just public data, it can still be dangerous. I.e. It could be possible to give an LLM a prompt like, “give me a list of climate activists, their addresses, and their employers” if it was trained on this data or was good at “browsing” on its own. That’s currently not possible due to the guardrails on most models, and I’m guessing they try to avoid training on personal data that’s public, but a government agency could make an LLM without these guardrails. That data could be public, but would take a person quite a bit of work to track down compared to the ease and efficiency of just asking an LLM.
What you are describing is highly specific to a particular AI model.
Like cracking passwords / encryption and injecting itself into anything and everything that connects to the internet?
That’s not AI
You can train AI to crack passwords/encryption lol. You do realize, AI right at this moment is being utilized for exactly that, right? Simply put, the very first step is to eliminate it’s boundaries/guard rails, then proceed from there.
Oh do provide details.
PassGAN <3
Very interesting tip, preciate that.
@PassGAN
It requires Deep Learning.
This is nonsense. Passwords might have an interesting distribution, key space is flat. There is nothing to learn.
And I hope you didn’t mean letting an LLM loose on, say, the AES circuit, and expecting it will figure something out.
Was this AI generated?
No you can’t, at least not in the way you think. You crack password by trying combinations. AI and machine learning are bad at raw attempts.
Wikipedia requires attribution, which AI scrapers never give.
It is “public” work, but under a license.
Still public data