As far as I know, the problem is still how to segment data flow from control plane for LLMs. Isn't that why we still can prompt inject/jail break these things?
I have previous CTO experience in the POS space, you would be right on both counts.
(till this day, I can walk into a shop, look at the POS screen and identify if its one of those visual basic/PHP/windows XP compatibility mode required stuff).
you may have a point, i.e. some mechanism to invoke a behavior that only a bot or LLM could do, that a human would not, e.g. click on this button now in a hidden div/transparent color or measure response time within page load.
the problem is that once this is found out, the circumvention is easy enough to program into bots/LLMS.
are we going to reinvent the voight-kampff test from bladerunner?!?
my g-shock frogman is 10+ years old, still sits by the window, charging itself everyday (solar powered), and it's the watch I would grab if the proverbial hit the fan. I'd grab my self-winding analog jlc watch too, just in case I needed it as a bribe (not sure which would be better as a bribe if that happens, maybe the g-shock!)
Sigh.
As far as I know, the problem is still how to segment data flow from control plane for LLMs. Isn't that why we still can prompt inject/jail break these things?
reply