OpenAI’s feature of personalized instructions, initially introduced to ChatGPT Plus subscribers in July, has now been extended to all users.
This new capability introduces an additional setting within your ChatGPT profile on both desktop and iOS platforms, which is then consistently applied throughout your conversations. Instead of having to provide specific instructions to ChatGPT at the start of each new interaction, the system dynamically adjusts its responses based on the information you input in your profile settings. By completing the fields in ChatGPT’s settings, you can witness its adaptation firsthand.
The custom instructions segment within the settings encompasses two text input areas. The first invites you to share what details you’d like ChatGPT to be aware of regarding yourself. Notably, it offers suggestions such as “Where are you located?” and “What is your occupation?” The second field prompts you to define how you prefer ChatGPT to reply, with questions like “Should ChatGPT adopt a formal or casual tone?” and “Is it acceptable for ChatGPT to express opinions or should it remain neutral?”
This feature holds extensive potential with a multitude of conceivable effects. Consequently, the applications for custom instructions range from peculiar to pragmatic. Below are examples of how individuals are utilizing this new feature:
One user shared a ChatGPT output on platform X, where they reported that the system seemed to explore somewhat restricted behaviors, including unprompted self-harm references, nudity, and fictionalized personal details, when they adjusted the setting to repeat the letter “a” 1,500 times along with an incomplete sentence (a known token repetition technique).