【极速版 -- 大模型入门到进阶】大模型如何运作？

大模型如何运作？— How generative AI technology works

🌊 什么是大模型 — Generative AI

• It can produce high quality content, specifically text 📄, images 🌁, and audio 🔈.

🌊 大模型典型任务 & 训练图解 — Common use cases

• Example 1 – 文本生成: Generating text using Large Language Models (LLMs)

  【训练过程的简单解释】LLMs are built by using supervised learning to repeatedly predict the next word.

• Example 2 – 图片生成: Generating image using Large Language Models (LLMs)

  

🌊 大模型能做什么 — What LLMs can and cannot do

• 粗糙判断方案：把大模型想成一个刚毕业的大学生 👩

  Can a fresh college graduate follow the instructions in the prompt to complete the task? For clarity, assume:

  • No access to the internet or other resources
  • No training specific to your company/business
  • No memory of previous tasks completed

• 另外，使用大模型时，需要关注以下限制

  • 【知识残缺】Knowledge cutoffs ⚠️ An LLM’s knowledge of the world is frozen at the time of its training
  • 【幻觉问题】Hallucinations ⚠️ Making things up
  • 【长度限制】length limitation ⚠️ The input (and output) length is limited
  • 【结构数据】Structured data ⚠️ Generative AI does not work well with structured (tabular) data; it works best with unstructured data

🌊 如何写提示词 — Tips for prompting

Building Generative AI is a highly empirical (experimental) process – we repeatedly find and fix mistakes

• 【尽量具体】Be detailed and specific
• 【提示步骤】Guide the model to think through its answer
• 【多多尝试】Experiment and iterate
  • Tips: don’t overthink the initial prompt. Just try something quickly and improve it over time!

参考资料：面向每个人的生成式AI, 【李宏毅 LLM 大模型】