ChatGPT Wants to Get Chummy, and That’s a Tall Order
ChatGPT has come up with its latest gimmick – “custom instructions”. It’s like a personalized greeting card that tells your AI how special you are, over and over again. Now, before you roll your eyes and jump on the ‘Yet Another AI Update’ bandwagon, hold onto your seats.
Remember Me, AI?
You see, the wizards at OpenAI are trying to make your life a teensy bit simpler, or at least, that’s the sales pitch. The aim is to cut down the typing rigmarole by helping ChatGPT remember bits about you. So the next time you’re conjuring up a moon-related fact for your third-graders, you won’t have to remind your virtual assistant that you’re not working for NASA. Handy, eh?
How To Access ChatGPT’s Custom Instructions
Since “Custom Instructions” are still in beta you will need to access it in that area. Just go to your account settings and switch it on as you did with plugins and the code interpreter.
The Gritty Details
Joanne Jang, one of the minds behind the model behaviors and product at OpenAI, seems rather excited about the prospect. According to her, these custom instructions are like a permanent reminder to ChatGPT about your particularities. Whether you’re cooking for a Brady Bunch sized family or scripting in a single language, it promises to tailor answers to your context.
However, let’s hold back the applause. The idea of crafting perfect instructions for the AI to give you impeccable responses sounds as whimsical as a unicorn picnic. As any tech-enthusiast will tell you, increasing complexity often leads to more room for error.
Expectations vs. Reality
During a demo, Jang inputted that she was a coder working in Golang. Then she asked for a cookie recipe, which apparently seemed too distant from coding for the AI to draw a connection. The response? A recipe formatted like Golang code. A stretch, sure, but it has its moments. When she then played it casual with a “Hey what’s up”, ChatGPT decided to leave Golang out of it, leading to a nod of approval from Jang.
The essence of launching this in beta is not just for the tech heads at OpenAI to pat themselves on the back. They actually want the model to figure out when and how to apply these guidelines – a smart move given the spaghetti bowl of possible outcomes.
The Future: A Dance of Give and Take
Ultimately, Jang imagines custom instructions as more dynamic, interactive, and adaptable to your needs. OpenAI is keeping a hawk eye on ensuring these custom instructions don’t override safety measures. So, any off-kilter requests like “how to commit the perfect crime” won’t be entertained. They’re also committed to scrubbing out any personally identifying information, so you can breathe easy.
At the end of the day, the aim is to make ChatGPT a more helpful, efficient assistant, taking notes about you while it serves. But as the saying goes, the devil is in the details.
