If you happen to construct it, people will try to interrupt it. Sometimes even the people constructing stuff are those breaking it. Such is the case with Anthropic and its latest research which demonstrates an interesting vulnerability in current LLM technology. Kind of when you keep at a matter, you may break guardrails and wind up with large language models telling you stuff that they’re designed to not. Like how you can construct a bomb.
In fact given progress in open-source AI technology, you may spin up your individual LLM locally and just ask it whatever you wish, but for more consumer-grade stuff that is a problem price pondering. What’s fun about AI today is the short pace it’s advancing, and the way well — or not — we’re doing as a species to higher understand what we’re constructing.
If you happen to’ll allow me the thought, I’m wondering if we’re going to see more questions and problems with the sort that Anthropic outlines as LLMs and other recent AI model types get smarter, and bigger. Which is probably repeating myself. However the closer we get to more generalized AI intelligence, the more it should resemble a considering entity, and never a pc that we are able to program, right? In that case, we might need a harder time nailing down edge cases to the purpose when that work becomes unfeasible? Anyway, let’s speak about what Anthropic recently shared.
