Frequently Asked Questions ​

Why can't GPT-4 or GPT-3.5 identify its own version? ​

If you ask GPT-4 "Are you GPT-4?", it will very likely reply: "I am OpenAI's GPT-3 model, and GPT-4 has not been released yet."

The reason behind this is that the training data for the GPT-4 API provided by OpenAI has a cutoff of September 2021. Once training is complete, the model's knowledge does not update automatically without retraining. Just as you couldn't answer questions about 2023 events back in 2021, the model's answers about its own version will inevitably be incorrect.

How do I tell GPT-3.5 and GPT-4 apart? ​

Try asking: "There are 9 birds on a tree. A hunter shoots one. How many birds are left on the tree?"

In 90% of cases, GPT-3.5 will answer 8 birds remaining. GPT-4 will most likely answer 0, because the rest flew away from the noise.

If this test is inconclusive, start a new conversation or try a few more times. Do not ask repeatedly within the same context.

What are Tokens? ​

A token is the basic unit GPT uses to process text. In short, a token can be a word, a part of a word, or a character in a specific language. Tokens convert input text into a format GPT can process. Roughly 1,000 tokens ≈ 750 English words ≈ 400 Chinese characters.

Each GPT model has a preset maximum token count. For example, GPT-3 supports a maximum of approximately 4,096 tokens. Note that this limit includes both input and output tokens combined.

What timeout value should I use? ​

• gpt-3.5: 120 seconds recommended
• gpt-4: 10 minutes (600 seconds) recommended
• Streaming output is strongly recommended — non-streaming responses are prone to triggering timeout
• For streaming output (typewriter effect), refer to the 🚀 JS SSE typewriter effect section in Quick Start