These LLMs are essentially just an open eval() as they are now. I feel like any attempt at obfuscation or post-facto mitigation of these kinds of exploits are doomed to fail in the long run.
That being said, with the amount of resources and talent being poured into AI research these days, I’m sure a more controllable version or evolution of these LLMs is already cooking up in a lab somewhere.
These LLMs are essentially just an open
eval()as they are now. I feel like any attempt at obfuscation or post-facto mitigation of these kinds of exploits are doomed to fail in the long run.That being said, with the amount of resources and talent being poured into AI research these days, I’m sure a more controllable version or evolution of these LLMs is already cooking up in a lab somewhere.