Are AI agents overrated and ultimately just hallucinations that are forwarded and used as input to hallucinate even more? I notice that people describe AI agents, but also the system prompts, as if they were talking to a human being. They use polite phrases, repeat themselves, some even claim that being polite to an AI created better results etc…
I say it works like this: An agent receives a task written in human-readable form with politeness and such. Then it creates some other human-readable statement and sends it to another AI. Every time there is some hallucination. If the end result is not totally off, the inventors say: it works.
It bugs me that I am yet again the only one to address this, everyone else in those low-quality videos on youtube praises ai agents just like the next open source model that now has even more parameters.
It feels like those people in praising videos like “Two minute papers” or similar, they are not even real people, at least they are pathologically autistic.