Back to Basics
Just like that, the news is in: GPT-4 has been officially unveiled earlier this week, leading some to believe that it outshines and leaves ChatGPT in the dust.
However, it's important to clarify that GPT-4 is not a replacement for ChatGPT; rather, it serves as the underlying language model upon which the conversational agent is based. This nuanced difference highlights the pervasive confusion that has arisen from the rapidly evolving world of generative AI technology.
From Dorking to Prompt Whispering
Amidst this evolving landscape, there is a burgeoning array of training courses and guides, with varying degrees of quality, that aim to help people make sense of this new discipline and become familiar with the subtle art of whispering to AIs.
At the core of this emerging field lies the art of "prompt engineering", a skill that resonates with the research technique of "dorking". This involves using specific queries across one or multiple search engines to procure the most accurate and relevant results possible. Prompts, as a reminder, are the words or phrases that guide the AI in generating text, images, videos, and even music.
Those with a hacker mentality are now delving into "reverse prompt engineering", whereby they task the AI with generating the prompts itself. While this method demands a certain level of finesse, it can prove to be highly rewarding, as the AI can leverage its own cognitive abilities to produce the most fitting prompts.
As you can see, the devil is in the details. Allow me to share a prompt written for DALL-E, a generative AI tool that enables the creation of images.
The precision and technicality of this prompt are immediately striking. It's evident that the author is, at the very least, an amateur photographer who has mastered the fundamentals of the art. And that's precisely the point I want to make.
When it comes to having ChatGPT generate content for you, do you want it to produce just any well-written material or compelling copy that sets you apart from your competitors? What makes your product or service unique, and how can ChatGPT highlight those features? What tone do you want to convey - casual, everyday, or formal? Are you looking to inject humor into the text, and if so, what kind of humor would be appropriate - sarcasm, irony, parody, or satire? On the other hand, if you want the writing to reflect a specific style, such as the style of Camus, it's worth considering what that entails. Does it mean emphasizing simplicity, clarity, and effectiveness? Or does it involve using metaphor and allegory to illustrate ideas?
Bear in mind that the caliber of ChatGPT's responses is intimately tied to the quality of the questions you pose. After all, ChatGPT essentially functions as an executor.
Human expertise comes first
While ChatGPT is indeed impressive on the paper, its effectiveness ultimately relies on the skill and expertise of its users. The more knowledge you have about your subject, the better you can generate precise prompts, ask relevant questions, engage in a productive dialogue with the chatbot, and most importantly, assess the quality and subtlety of its responses to push yourself and the AI to excel. It is also important to note that, at times, ChatGPT may make mistakes.
As copywriter Marie Gaymard pointed out to me in a recent LinkedIn comment, "It's true that without any awareness of your own editorial signature, it seems difficult to guide the AI with precision. The same goes for visuals where we can clearly see the difference between prompts from skilled creatives and those from newbies." It's clear that AI won't replace humans, but those who can use it effectively will certainly outpace those who can't, or worse, those who mistakenly believe they're using it well and inundate us with mediocre content.
In this new context, it seems more essential than ever to exercise critical thinking. But that's not all. In the age of selfpressionnism and the explosion of big data, we must also learn to develop a form of critical ignorance to filter sources and identify the most relevant information.
The proliferation of generative AI is set to deepen existing social inequalities. With current free language models taking a broad-brush approach to content processing, access to fine-tuning - training models on specific thematic data sets - has become an increasingly critical factor in unlocking the full potential of AI. Without the skills to fine-tune, many individuals and businesses will struggle to fully harness the power of these tools. In the same way that one wouldn't expect a novice driver to handle a Ferrari, it's unreasonable to expect individuals without the necessary skills and resources to fully leverage the potential of (advanced) AI models.
In other words, it's essential to return to the fundamentals. To have a solid base of general theoretical knowledge as well as specialized, sector-specific knowledge in order to guide and use AI effectively. By investing in your own education and lifelong learning, and developing informed reasoning, you ensure that you're able to navigate the complex world of AI with relevance.
Humanoid GPTs on the Horizon
You can be sure of one thing: ChatGPT and its counterparts are just the visible tip of the iceberg. As Microsoft 365 announced the integration of ChatGPT (and more) into Word, Teams, Excel, and other tools, it's only a matter of time before generative AI enhances the intelligence of robots, drones, and even humanoids. This technology will gradually infiltrate our daily lives, almost seamlessly.
However, there's more to the story. As generative AI captures global interest, a parallel revolution emerges with the advent of affective computing. This technology aspires to equip machines with the ability to recognize, interpret, process, and even mimic human emotions, delivering a more personalized and empathetic user experience. With the Affective Computing Market valued at USD 37 billion in 2021 and projected to reach USD 622 billion by 2030, growing at a CAGR of 36% from 2023 to 2030, it's clear that this field is gaining momentum. Affective computing complements generative AI, prompting us to wonder: are we already on the cusp of transitioning from "prompt engineering" to "humaneering"?
My Go-To Resources
If you're looking to enhance your skills in prompt engineering, I recommend checking out the open-source community platform Learn Prompting. It offers comprehensive coverage of the topic with varying levels that demonstrate the nuances of prompt engineering.
Another great resource is the DALL-E promptbook, which provides practical advice and concrete tips. It emphasizes the importance of mastering the fundamentals, such as artistic movements, creative styles, photographic techniques, and the emotional language required.
To stay up-to-date with the rapidly evolving world of generative AI, I highly recommend subscribing to the newsletters Towards AI and Ben Bites daily. Both offer expertly curated content on everything related to generative AI, including new tools, must-read articles, and upcoming events.
The report on the future of content, co-authored by myself, Benoit Zante, and Quentin Franque, which is regularly updated.
AIRPRM is an extension that modifies ChatGPT's homepage. It offers a wide range of prompts classified by theme, although not all of them are top-notch. One feature I appreciate is the ability to save my own prompts.