Prompt engineering: The dawn of the AI whisperer

In many ways, it’s like whispering a spell. Say it slightly wrong, and it won’t work.

PREMIUM

Anyone who has used an Artificial Intelligence (AI) language or art program (and really, what are you waiting for if you haven’t?), knows how vital the right query can be. With the correct cues, AI art can be indistinguishable from the real thing; mystifying concepts are made easy enough for a six-year-old to understand; fresh code appears as if by magic.

In these early days, the right cues also act as an education program. Like an encouraging teacher, the cues nudge the program to learn better, gain nuance, offer context. Which is why one of the first new jobs created by AI is the prompt engineer.

In addition to crafting conversations, prompt engineers are tackling ethical concerns and customising AI solutions across industries, says Texas-based data scientist Shubham Saboo, 26, author of GPT-3: The Ultimate Guide to Building NLP Products with OpenAI API. Saboo is also head of developer relations at Tenstorrent, which builds processors for AI.

He recently compiled a Twitter thread on the online resources that can help one learn about this new profession. There are modules on offer on upskilling platforms. Coursera’s Prompt Engineering for ChatGPT course is free. Udemy’s The Complete Prompt Engineering for AI Bootcamp offers about 62 downloadable resources that cover limitations and best practices too. The course costs ₹799.

There’s an online marketplace called PromptBase that allows people to hire prompt engineers or sell their prompts.

And, already, companies such as Anthropic (an American AI research and safety start-up) and Klarity (an AI-based document-review service) are offering prompt engineers packages ranging from $130,000 to $375,000, according to posts on company websites and on LinkedIn.

A background in machine language or AI isn’t necessary, but can come in handy, depending on the AI tool, says Anna Bernstein, 29, a New York-based prompt engineer with Copy.ai, an AI-powered content-creation platform. She studied English literature, worked as a copywriter, and is a published poet. She’s been with Copy.ai since 2021 and got some initial guidance from a machine-language consultant.

Now, she spends her days researching AI “behaviour” and developing new toolkits to help Copy.ai’s content creators use AI more effectively.

Her background in literature is crucial, because a good prompt engineer must know what the right output looks like. “My current role brings together my obsession with language, love for the craft of writing and this mad need to get things right,” she says.

As Andrej Karpathy, Tesla’s former chief of AI, put it in a tweet in January: “The hottest new programming language is English.”

Soon, however, the rudimentary principles of prompt engineering will likely be widely known; just as most people have learnt (often without realising it) how to reliably use the Google search engine.

“As AI models improve, the need for complex instructions will become obsolete. As AI becomes better at following instructions, using it will be as simple as talking to a human,” Saboo estimates.

What will the prompt engineers pivot to then? Ethics and bias management, AI education, copyright protection, prevention of plagiarism, and policy research, says Saboo.

They could become part of the frontline against the illegal and irresponsible use of AI. Which will involve spells of a whole other kind.