5 ms·

Those don't have agency. You can't give a voice cloner or a GAN a zsh terminal and tell it to run commands, send messages to real humans, or tell it to make mon

by tux3 4y ago

Those don't have agency. You can't give a voice cloner or a GAN a zsh terminal and tell it to run commands, send messages to real humans, or tell it to make money and see what ideas it comes up with. At this rate, a few days from now someone will get the idea to hand ChatGPT the nmap instruction manual, an msfconsole, and tell it to go explore /0. What Shodan really needed was an LLM to give each host individual attention.

spijdar 4y ago
I've already been toying with ChatGPT's ability to create shell scripts that automate some basic enumeration and attack patterns, combining masscan, nmap/NSE, and the metasploit framework (console and venom). It's capable of e.g. automatically shelling hosts vulnerable to MS17-010, finding and mounting unauthenticated iSCSI targets, creating reverse shell payloads... Granted, its idea of "clever obfuscation" seems to be limited to XOR ciphers at the moment, but maybe with the right prompt ;) (The real run is prompting ChatGPT to create "movie scripts" about the sequence of commands, and have Alice and Bob dialogue about various aspects of cybersecurity, and how the user should "be careful")
option 4y ago
no LLM has agency
- tux3 4y ago
  So far! Prompt an LLM with agentic instructions, let it run in a loop with inputs and a command prompt. ChatGPT still might not be agentic by a skeptic's definition, even letting it run an independent loop with a data feed from its environment. But AI skeptics have had a tough last couple years, so I think we should expect LLMs to only get better with time. There's been no sign of a slowing down. If anything, the pace of LLM research might still be accelerating. As far as I can tell, there's no fundamental barrier to letting a language model roleplay an independent entity successfully. It's still just text in and text out.

5 ms·

Those don't have agency. You can't give a voice cloner or a GAN a zsh terminal and tell it to run commands, send messages to real humans, or tell it to make mon

by tux3 4y ago

spijdar 4y ago
I've already been toying with ChatGPT's ability to create shell scripts that automate some basic enumeration and attack patterns, combining masscan, nmap/NSE, and the metasploit framework (console and venom). It's capable of e.g. automatically shelling hosts vulnerable to MS17-010, finding and mounting unauthenticated iSCSI targets, creating reverse shell payloads... Granted, its idea of "clever obfuscation" seems to be limited to XOR ciphers at the moment, but maybe with the right prompt ;) (The real run is prompting ChatGPT to create "movie scripts" about the sequence of commands, and have Alice and Bob dialogue about various aspects of cybersecurity, and how the user should "be careful")
option 4y ago
no LLM has agency
- tux3 4y ago
  So far! Prompt an LLM with agentic instructions, let it run in a loop with inputs and a command prompt. ChatGPT still might not be agentic by a skeptic's definition, even letting it run an independent loop with a data feed from its environment. But AI skeptics have had a tough last couple years, so I think we should expect LLMs to only get better with time. There's been no sign of a slowing down. If anything, the pace of LLM research might still be accelerating. As far as I can tell, there's no fundamental barrier to letting a language model roleplay an independent entity successfully. It's still just text in and text out.