DailyPost 2497
CHAT GPT – CUSTOM INSTRUCTIONS FEATURE
Our inability to keep improving upon the utility of any pathbreaking system makes it stagnant. The current wave of large language models starting with the revolutionary ChatGPT have a different story to say. The iterative or what we can say, new use cases are being created by the dozens but there are not many takers. They seem to be more than happy with the standard model and answers they derive out of it. ChatGPT being a product of OpenAI is committed improving on what they created and make as utilitarian as possible. Each of the iterations are bound to be pathbreaking.
This is exactly why this AI stage is set; where we witness momentous changes many a times, but percolating it down beyond the generic level rarely happens. Mass AI would then remain much farther than we think. However much we claim to the contrary, continuous learning does not seem to be our fate. It is a trait which most of us would love to dispense with. But as we move forward, given the exponential changes in the AI landscape, if we continue with this trait, we would be the losers. A new feature release on ChatGPT is changing the game. This feature is called custom instructions.
This is start of personalization of this great moving away from just having a context of a continuing conversation / dialogue. A blog of 20th July 2023 from the creators says that they are introducing custom instructions so that the user can tailor ChatGPT to better meet his needs. ChatGPT will consider your custom instructions for every conversation going forward. The instructions so to say gets coded and you don’t have to repeat your preferences. For example, a teacher crafting a lesson plan no longer has to repeat that she is teaching third grade, the answers would be accordingly fine-tuned or a coder preferring efficient code in a language that’s not python.
If the output format is also made a part of custom instructions, precision of different kind comes as the output. Perfunctory text can also be avoided in this manner. You call these as system instructions. The personalization of AI now starts in a very serious manner, and this seems to be the future of LLMs. ChatGPT can be broken much more easily now with custom instructions, and can also override general instructions of the platform. With custom instructions, you can create * precision prompts for Midjourney* out of ChatGPT. The playground feature of ChatGPT also has a system column which with the right input, helps in getting the most appropriate output. You can save this as preset. Lots of other features are also provided here. It would not be long when custom instructions would start getting shared on the internet.
IF WE DON’T DON OUR LEARNING CAPS, WE MIGHT END UP AS AI UNTOUCHABLES.
Sanjay Sahay
Have a nice evening.