alt Hacker NewsIts incredibly depressing that LLMs have made so many facets of life strictly worse. Everything is absolutely inundated with slop
I'm mildly hopeful that when text watermarking comes in, we'll be able to automatically filter out all the LLM crap that gets generated currently. It feels like we're in the very early days of email becoming a tool used for mass spam, while we figure out exactly how to deal with that
Replies
gedy • today at 4:00 PM
Relying on in person relationships and communication are vastly better if digital channels get saturated with noise.
➕ show 1 reply
watermarking will only work if it's embedded in the model (i.e. frontier/paid offerings). The rate at which open-source/open-weight models are advancing I think watermarks are probably only a temporary solution right? Fragmentation of the models available, or having more models that don't contain watermarks available will probably make this technique useless in the future I would think?