Whispering in the ear of artificial intelligence, a profession of the future
If, while testing ChatGPT, you felt smug with conversational artificial intelligence (Is) of this type,…
If, while testing ChatGPT, you felt smug with conversational artificial intelligences (AIs) of this type, your skills may be in demand. In the jargon of this emerging technology , the professionals occupying this function are called “prompt engineers”, the word “prompt” meaning the way in which one initiates and directs a conversation with a machine of this type. There are still very few of them, but their job description is as futuristic as the generative AI they are working on.
Their role is to discuss with it in their mother tongue (in “natural” language), and therefore without necessarily knowing how to code or master the computer science of artificial intelligence. The goal is twofold: to guide the conversational robot in its future interactions – and therefore in particular to teach it to avoid making inappropriate comments -, but also to find the best ways to obtain the best possible answers.
Educate and train
For the first function, it is a question of registering in the system of the conversational robot – even if this is done in natural language – specific instructions giving contextual elements for the conversation to come, and that the user does not see . “For example, we ask the robot to be polite or we tell it that it is an astrophysics teacher responding to its students so that it is relevant”, explains Amélie Chatelain, “prompt engineer” at LightOn a French start-up that develops a ChatGPT-style language model.
The second function aims to train generative AI. “This is the main role of prompt engineers,” explains Amélie Chatelain. Concretely, for specific uses of customers of companies like OpenAI, the prompt engineer will seek the most effective ways to respond to a given problem.
If coding is not necessarily necessary, knowing how to do it can help to automate interactions with the robot in order to educate or train it in natural language. In any case, scientific training seems to be preferred, the emphasis being on logic.
This job sometimes arouses a certain skepticism – how to guide a robot whose multiple answers can surprise their designers themselves? – or is considered only temporary until the artificial intelligence itself learns to do without them! In the meantime, he seems to be paid very well. An announcement from generative AI company Anthropic recently offered a salary that could exceed $300,000 per year.