By Adam Pease
OpenAI Brings Custom Instructions to ChatGPT
OpenAI recently announced that it would be adding a new feature, custom instructions, for ChatGPT. This innovative functionality is designed to better align ChatGPT with user needs, and will be initially available for the Plus plan, with a broader roll-out scheduled over the coming weeks.
This blog discusses the news and its implications for the generative AI market.
What Are Custom Instructions?
Custom Instructions is a response to user feedback concerning the repetitive aspect of starting each ChatGPT conversation from scratch. The OpenAI team has interacted with users from a variety of countries and learned about the importance of aligning AI behavior with diverse user contexts and individual requirements.
With this new feature, users can define their preferences or requirements, and ChatGPT will take these into account in all future conversations, eliminating the need to repeat the same information in every session. For instance, a teacher can specify that they’re teaching 3rd grade science, and this preference will be remembered by the model in subsequent conversations.
Similarly, a programmer can define a preference for efficient code in a specific non-Python language, or a parent can set a requirement for grocery lists to include quantities for six servings.
What’s Next for Generative AI?
The announcement of custom instructions suggests that the application capabilities of generative AI are just now coming online. It solves a big problem faced by generative AI users—the need to continuously re-prompt and re-prime language models in new sessions given new context.
Equipped with custom instructions, users can define a predetermined template of behavior that they can revisit in future sessions, eliminating repetitive work and preserving best practices throughout an organization.
Generative AI is still very new, and organizations are still learning how to leverage it most effectively and to build applications around it well. Custom instructions represent just one highly-desired feature that is coming online for language models, and it will no doubt expand the utility of these tools in ways that we can expect to continue growing over time.
One sign of this is the way that custom instructions can be integrated with another existing ChatGPT feature, plugins, to construct persistent instructions that will integrate with outside systems and APIs.
Bottom Line
The race to more sophisticated generative AI is not slowing down. While OpenAI’s announcement marks just one critical emerging feature for LLM-based applications, it suggests to Aragon that organizations should expect the capabilities of these tools to continue to grow in more flexible and user-friendyl ways in the near future.
How Can You Use AI to Help Combat and Survive these AI-Powered Cybersecurity Attacks?
Tune in for our complimentary webinar on August 17th, 2o23 | 10 AM PT / 1 PM ET
Cybercriminals are aggressively weaponizing artificial intelligence (AI) to launch increasingly effective cyberattacks against organizations. These cybercriminals are using AI to launch sophisticated and stealthy cyberattacks, such as creating realistic deep fakes, generating malware that can evade detection systems, creating convincing phishing emails, or identifying and exploiting vulnerabilities in real-time.
In this webinar, you will learn how AI can help you fight fire with fire to combat and survive these AI-powered cybersecurity attacks.
Some key areas we’ll cover:
- How cybercriminals are changing the game with AI?
- What solutions are available to combat these threats?
- How emerging AI technologies will transform cybersecurity?
