Hello, I noticed with all models that when you add training, especially with longer training, GPT-3 focuses more on the more recent training or the training closer to the end. When in doubt, it usually will repeat something from the most recent chunk of training. It can be rather annoying when tryin…

Location of information in training

Topic		Replies	Views
When processing a text: prompt before it or after it? Prompting chatgpt	13	8093	December 14, 2023
Repeated questions improve quality of assistant's responses Prompting assistants-api	4	191	April 29, 2025
Does gpt-3.5-16k instruction following ability decrease as the system prompt gets larger? Prompting gpt-35-turbo	4	1285	December 14, 2023
Same prompt giving required output and then missing details when reprompted again Prompting gpt-4 , api	5	822	April 9, 2024
Finetuning not working? Prompting	8	2568	December 24, 2023