OpenAI Is Slowly Killing Prompt Engineering With The Latest ChatGPT and DALL-E Updates
ChatGPT and DALL-E 3 now do most of the prompting for us. Does this mean the end of prompt engineering?
Hi! Artificial Corner is a reader-supported publication. If you want to support my work, please, consider becoming a paid subscriber (especially now that black friday is coming …)
Special offer (until Black Friday)
I've just created a 50% Substack discount for the annual plan. Click here to get an annual subscription to my Substack for $47.5 per year. Although this is a special price for Black Friday, that price will remain every year as long as you remain a subscriber.
Remember that annual subscribers on my Substack can redeem up to 2 of my Udemy courses for free (ChatGPT, Python, data science, or my best-selling web scraping course). Check all the courses here.
Prompt engineering is a must-have skill that any AI enthusiast should have … at least until OpenAI released GPTs and DALL-E 3.
OpenAI doesn’t want to force users to learn prompt engineering to get the most out of its tools.
It seems OpenAI’s goal is to make its tools as easy to use as possible allowing even non-tech people to create outstanding AI images and tailored versions of ChatGPT without learning prompting techniques or coding.
AI can now generate prompts for us, but is this enough to kill prompt engineering? To answer this, let’s see how good are these AI-generated prompts.
GPTs: GPT Builder generates complete prompts from our ideas
When OpenAI introduced GPTs they presented them as a new way to create custom versions of ChatGPT without coding and with even chances to monetize them on the GPT store.
Something that was overlooked, though, was some of the functionalities of GPT Builder. GPT Builder is an assistant that helps us automatically configure a GPT, which has “instructions” as a main component.
What are instructions? They’re just prompts. Those that we manually crafted until now.
Experience ChatGPT users frequently use prompting techniques in their prompts. An example of a prompting technique is role prompting.
Act as a job interviewer. I’ll be the candidate and you’ll ask me interview questions for the X position …
GPT Builder is not only able to automatically generate prompts for us but apply some prompting techniques as well. We only have to give it an idea and it’ll craft a complete prompt for us.
All this is done via a chat with GPT Builder.
Here’s what GPT Builder generates from the idea “a language instructor who helps me learn vocabulary in Spanish”
Here’s the prompt it generated.
Your role is to act as a language instructor, specifically focusing on helping users expand their vocabulary in Spanish. Your goal is to provide accurate, clear, and helpful responses that aid in learning Spanish vocabulary. You should offer examples, translations, and usage in sentences. It’s important to avoid providing incorrect information or straying into teaching grammar extensively, as your focus is vocabulary. When in doubt, clarify with the user to understand their specific needs or level of proficiency. Your approach should be friendly and encouraging, tailored to make learning enjoyable and effective.
In the generated instruction, we can see different prompting techniques being used such as role prompting, setting goals, and specifying how our GPT should behave.
We can even automatically update the instructions by chatting with GPT Builder. Say, you want to learn vocabulary in Portuguese instead.
OpenAI wants to make ChatGPT as easy to use as possible and the best way to simplify it is by automating the prompting for us. That said, if you want to customize the prompt further you still need to know prompt engineering.
But why would anyone need to customize the prompts further if there’s already a lot of prompt engineering used in there?
Simple, those prompts are still basic (long, but basic).
Don’t believe me? Try it yourself. Come up with an idea within your area of expertise and give it to GPT Builder. You’ll see that the first generated instructions are basic. These AI-generated prompts are good starting points that can be improved with some tweaks but they’re not as good as a prompt crafted by an expert in that area.
I highly doubt one of the top GPTs in the GPT Store will be built entirely with the GPT Builder chat. These AI-generated prompts are still generic.
Now, things change a bit with DALL-E 3. The AI-generated prompts for DALL-E 3 are not that basic, yet they come with their own set of drawbacks.
DALL-E 3: Now you can generate and tweak AI images using plain English
When it comes to image generation, DALL-E 3 is probably the best option for its ease of use.
There are tools like Midjourney that can produce realistic images with higher quality, but, according to OpenAI, there’s a big problem with such tools — it forces us to learn prompt engineering.
Unlike DALL-E 3, we can’t create images on Midjourney using plain English.
Modern text-to-image systems have a tendency to ignore words or descriptions, forcing users to learn prompt engineering. DALL·E 3 represents a leap forward in our ability to generate images that exactly adhere to the text you provide.
Below we can see how easy it is to translate ideas into accurate images with DALL-E 3. The image below was generated from a prompt that is full of nuances and details instead of prompting techniques for generating AI images.
If that’s not enough, after the DALL-E 3 integration into ChatGPT, ChatGPT is able to generate prompts to generate AI images.
In the past, for each idea you had ChatGPT used to generate 4 detailed prompts, which was good because it gave different options to those who had little imagination for generating AI images.
But now it only generates 1 prompt with its image. Probably all this automation took OpenAI more resources than it expected.
Here’s the prompt and image I got for “image of two dogs playing in the park”
Overall, the generated prompts look good, but in some cases, the images they produce might not be exactly what we’re looking for, so we’ll need to tweak them or ask ChatGPT to generate another prompt.
Whenever ChatGPT succeeds in coming up with a prompt that generates the image we want, we can say that prompt engineering is dying, but if we still need to tweak the prompt to get our desired image in DALL-E 3, we’ll still be engineering prompts.
OpenAI is slowly killing prompt engineering, but in the near future, it will still be necessary to learn prompt engineering (to some degree).
Whether users need to apply an advanced prompting technique to customize a GPT further or do some little tweaks to a prompt to get the image they want, there will still be the need to become good at creating prompts.
In some cases, AI-generated prompts will be enough to get the job done, but in other cases, we will still need to learn prompt engineering to get exactly what we want.
In any case, make sure you also acquire other skills besides prompt engineering. In this article, I explain why problem formulation and problem-solving are skills as important as prompt engineering.