The vast majority of content online is now machine generated. Even before the advent of LLMs, most of the information posted online from the late 2000’s onward was either malicious or insane. Almost everything you see online is part of information warfare. Academic journals, legacy media, social media posts, video content, podcasts, etc. Very little information was honest, but with the power of automation now it is difficult to even find a blog that is written by a human at all.
Taking this site as an example, our view stats measured in page hits have gone off the charts since LLMs have been in the mix, but I strongly suspect that the number of human readers has actually gone down substantially. Most humans just don’t read anymore. They don’t do web searches, either. They ask an LLM to summarize, which sometimes works, but the LLMs have themselves ingested so much malicious information that their summaries can simply propagate more malicious information.
With this in mind, it is important to define “nonsense” and “antisense”.
Nonsense
Nonsense is innocent. It is noise generated from misunderstanding. Nonsense is not generated with an agenda or deception in mind, it is just a (sometimes non-trivial) set of mistakes, and mistakes can be corrected. There is enough signal in the background to override nonsense over time. There isn’t a person alive or who has ever lived who wasn’t mistaken about something, learned more, and corrected themselves.
Antisense
Antisense is malicious. It is information that is crafted specifically to bind to and silence signal. It masquerades as sense, and will typically propagate more quickly and widely than nonsense specifically because it was crafted according to a set of logical priors, so it feels very much like sense.
In summary, nonsense is inevitable entropy, but antisense is a malicious payload.