OpenAI-365

GPT User Guide

Friendly Reminder

We recommend using a combination of GPT-4 and GPT-3.5 for the best results and cost efficiency. GPT-4 excels at handling complex tasks, while GPT-3.5 is perfect for managing a large number of simpler tasks. This way, you can harness the full power of GPT-4 while keeping costs at a fraction of the official price.

Why is it so important to combine their use?

Based on our experience with users processing billions of tokens, when using GPT-4 for complex and critical tasks:

  1. GPT-3.5 can fully handle the remaining less complex tasks, with a response speed that is four times faster than GPT-4, and a token price that is only 1/20 to 1/30 of GPT-4 (since GPT-4 is 20-30 times more expensive than GPT-3.5).
  2. If you still want to use GPT-4 exclusively, make sure to avoid adding too much irrelevant context in the conversation. Doing so will not improve understanding of the context and will consume more tokens.

Why can't GPT-4 or GPT-3.5 recognize their own version?

First of all, if you ask GPT-4: Are you GPT-4? It is very likely to answer: I am OpenAI's GPT-3 model, and GPT-4 has not been released yet.

The reason behind this phenomenon is that the GPT-4 API provided by OpenAI uses training data up to September 2021. Once the model is trained, the knowledge it contains does not automatically update unless it is trained again. Just as you cannot answer in 2021 what your first meal in 2023 was, your answer would inevitably be incorrect.

How to distinguish GPT-3.5 and GPT-4?

You can try asking, "There are 9 birds on a tree. A hunter shoots one bird. How many birds are left on the tree?"

  • In 90% of cases, GPT-3.5 will answer that there are 8 birds left;
  • while GPT-4 will most likely answer that there are 0 birds left, because the others were scared away.

If this question does not work, you can create a new context or try again several times. Please do not ask continuously in the same context.

What are tokens?

Tokens are the basic units that GPT uses to process text. In short, a token can be a character, word, or specific character in a particular language. They are responsible for converting input text data into a format that GPT can process.

  • Effect: The number of tokens affects the ability of the model, such as understanding complex semantics, expressing rich content, and efficiently processing long text.
  • Limitations: However, a higher number of tokens means larger computing resource requirements, which may result in slower processing speed and increased memory requirements.

Each GPT model has a preset maximum number of tokens that it can handle. For example, GPT-3 allows a maximum number of tokens to be processed, which is approximately 4096. It is important to note that this number includes all input and output tokens.

What is context?

When GPT is used for text generation, it needs to consider all previous input text context to generate coherent and meaningful sentences. As the input context increases, the text generated by GPT becomes more coherent and accurate. For example, if an entire article or paragraph is used as input, GPT will be able to generate natural language text that conforms to the context coherence. Therefore, the more context accumulated by GPT, the higher the accuracy and coherence of the generated text will be gradually improved.