2026-07-08 # LLM Manifesto

open

LLMs, or Large Language Models, are pre-trained neural networks. They are giant math equations that predict the most likely token - or atomic unit of language - to follow some input. They are trained off of vast amounts of human-written text. These models are often further fune-tuned to predict text that resembles a two-person chat conversation, most commonly for use as a chatbot.

ijo

• Humans act for themselves and are wholly responsible for their consequences.
• Large Language Models are tools and nothing more. Tools are not responsible.
• Large Language Models have no feelings, no inspiration, and no compassion.
• LLMs are trained on human text. Thus, they mimic the same flaws as humans.

pini

LLMs are not magic. LLMs must not be treated the same way we treat humans. LLMs must not control us - we must control LLMs. LLMs cannot fully replace the work of humans, though humans may often work more effectively with LLMs. LLMs are often not the most effective or efficient tool for a job. LLMs are nothing more than the sum of their training data and sampling algorithms.