OpenAI has unveiled a major update to ChatGPT that provides users much more control over the conversational AI. The new “Custom Instructions” feature, currently in beta, lets you give ChatGPT specific prompts that it will remember and apply across all chats.

This upgrade represents a huge leap forward in functionality. Previously, ChatGPT would treat each conversation as independent, with no memory or context carried over. But with custom instructions, you can essentially define guidelines and priorities for how ChatGPT should behave that will persist in every interaction.

To use the feature, Plus subscribers can insert two prompts up to 1500 characters each in the settings tab under “Custom Instructions.” For example, you could provide guidelines like “Act as an assistant that is honest, harmless, and helpful” or “Do not make speculative claims without credible sources.”

According to OpenAI, the custom instructions will shape ChatGPT’s responses across the board to align with the provided directives. This allows much more control compared to relying solely on the AI’s original training.

Currently in beta, custom instructions are available to Plus subscribers everywhere except the UK and EU. OpenAI states that it is coming soon to free users as well.

In addition to custom prompts, ChatGPT’s underlying GPT-4 model has had its message cap doubled from 25 to 50 exchanges every 3 hours during a chat. This allows longer, more complex conversations before timeouts.

Together, these upgrades provide users far more power to mold ChatGPT into an assistant suited for their specific needs. As the AI grows smarter, customization and guidance will likely be key to productive human-AI collaboration.