“The original GPT-3 model had a context window of 2,000 tokens.”