Bolting a giant language model into your app so it stops being dumb.
means Embedding a large language model into an existing product or system to handle text understanding, generation, or reasoning tasks without building the model from scratch.
from Born from the explosion of accessible LLM APIs (OpenAI's GPT-3 in 2020, then 4, Claude, Llama) making it suddenly cheaper and faster to rent intelligence than build it. Before that, companies built NLP pipelines like they were assembling IKEA furniture. Now they just plug in.
github copilot — Integrated Codex (later GPT-4) into VS Code; 1.3M paid users
slack ai — Embedded Claude into workspace; summarizes channels automatically
stripe radar — Uses LLMs to flag fraudulent transactions; reduced false positives