3 ms·

AI LLMs are just curating what content to train on vs indexing the www. It’s actually even more gatekeeper-y

• visarga 3y ago
  What exactly does gatekeeping mean in the context of LLMs? One has several options: you could directly use a search engine, employ an LLM equipped with a search engine plugin for queries outside the training set, or personally fine-tune a LoRA with your specific data in a matter of hours. We have easy, moderate, and complex solutions at our disposal. Rather than gate keeping, LLMs actually compress the search index into a manageable size, enabling easier sharing. While it's impossible to download the entirety of Google, you can conveniently download a LLaMA, promoting both privacy and autonomy. This marks a departure from the days of having every click and search keyword tracked. Perhaps the advent of LLMs and Generative Image models by 2030 could allow us to cut the cord and navigate a virtual internet, have a room of our own to dream and be free, essentially functioning as an augmented imagination where we set all the rules. Local LLMs solve the old problem of accessing information without leaking out what you are looking at, something that bothered me every since BitTorrent and copyright wars.