Researchers from Adversa.AI have discovered an issue
While it might be true that LLMs could help fuel something like that, this is coming from an “AI agent” security company and is just an ad for their platform. I also strongly dislike the use of the whole “agent” bs which makes it sound like LLMs can do more than they actually can.
it’s the right terminology though, an agent uses an LLM, so yes, it has more agency than a simple LLM that can just reply to a prompt.
Depends on how you’re defining agency. When it comes down to it, it’s all I/O with no human-style agency, which is what “agentic AI” implies to your average stock market investor. It’s a way to imply AGI without actually getting to AGI, a CYA move from slop companies.
This is not philosophy. An agent can do more than a model on its own because it’s a loop, therefore, has a higher level of agency. AGI is not a requirement nor an implication.
A “loop” does not equate to agency. I’m not even saying this in just philosophical terms, I’m saying that even with that “loop”, it’s not agency in any sense, technical or philosophical. It’s a marketing term for the masses, not something that LLMs actually have.
Hardly a marketing term. Agentic systems is a term in computer science long before LLMs. It’s agency, maybe not the kind you and I have, but it is.
I suppose, but does the general public know that? I doubt most do. I’d continue to argue that the slop companies know that the public has no idea of the difference and use that to their advantage. To most, it implies that it’s more than just an LLM firing based on conditions met (basically just automation) or more than just an LLM ingesting results from a prior run. That is why I hate the term, and what I’m trying to say. It’s all part of the calculated grift; it’s another term that the slop companies abuse for their own gain (especially in marketing, line goes up if you can use “AI agents” in your marketing, even if no one is buying).
