The rise of powerful large language models (LLMs) and other AI models have created a whole host of new career paths and opportunities.
Chief AI Officers and machine learning engineers are more in demand than ever before. Now prompt engineers are the latest role attracting a lot of buzz.
While some headlines tout six figure salaries available to prompt engineers without a STEM degree, the situation is a little more complex on the ground. Let’s look at what a prompt engineer does, the path to becoming one, and what sort of salary you can expect.
What is prompt engineering?
Large language models are supercharged autocomplete machines. Their neural networks take an input and try to predict the most appropriate next string of text. They are also static, in that each prompt will run through the whole neural network to garner a response. These raw features are abstracted away using techniques like fine-tuning, adjustable parameters that allow you to control the length and randomness of the output, and chatbot interfaces like OpenAI’s ChatGPT.
As a result, the input you give to an AI model is incredibly important. Take ChatGPT. The text you enter into the chat box is only a small part of the overall input the LLM receives. As well as your message, there’s also a set of secret instructions from OpenAI, all the previous messages in the conversation, a system prompt that reminds the application it is “a helpful assistant”, and potentially a whole lot more we don’t – or can’t – know about.
The emerging field of prompt engineering focuses on crafting an effective input prompt to get the model to respond how you want it to. It isn’t just a matter of feel or trial and error, but the subject of academic research and scientific techniques like chain-of-thought prompting and few-shot prompting.
Prompt engineer roles and responsibilities
A prompt engineer is responsible for developing, testing, and deploying the input prompts used to generate output from AI models. They may also be responsible for deciding which AI models to deploy, what configurable parameters are set, and how they interact with other systems using APIs and other integrations.
Where a prompt engineer fits in the overall organizational structure varies significantly. In small organizations where the engineering department has been offshored or out-sourced, it can be a senior role, equivalent to a less technical chief AI officer. In larger or more AI-focused organizations, it may just be a synonym for AI engineer or machine learning engineer. Zach Bartholomew, vice president of product at Perigon, says that anyone on their product and engineering teams could be considered a prompt engineer – even if that isn’t necessarily their job title.
While some prompt engineers in some organizations have a journalistic or liberal arts background, the vast majority of people deploying AI models on a product or engineering team have a technical background in subjects like data science, machine learning, and artificial intelligence.
What does a prompt engineer do?
As AI models continue to be widely adopted, understanding how to deploy them most effectively is the key skill of a prompt engineer. The specifics of the role depend on how AI models are used by an organization. Some of the things a prompt engineer will have to do are:
- Evaluate and select the appropriate AI model for different use cases based on performance, price, and other criteria.
- Develop, test, and optimize effective input and system prompts.
- Work with the dev team to deploy AI models within a product or internal tools.
- Troubleshoot and fix any issues with AI model responses.
- Manage and update data and knowledge sources so they can be used by AI tools.
- Keep up to date with the latest developments in AI models.
While prompt engineering can be a front end skill, in most organizations prompt engineers will also be working on the back end, configuring and deploying the AI models under the hood.
Prompt engineer vs AI engineer: what’s the difference?
The line between prompt engineers and AI or machine learning engineers is incredibly blurry. Bartholomew feels that the role of a non-technical prompt engineer will be “fleeting at best” as prompt engineering becomes a skill that is necessary across the entire organization, rather than a niche specialty. Any engineer deploying AI models will by necessity be a prompt engineer, in addition to whatever their existing role is.
Similarly, most job listings for prompt engineers currently require technical degrees, skills, and experience. Many seek computer science, data science, or computational linguistics qualifications, technical skills like computer programming and data modeling, and experience with cloud technology stacks and specialized tools like LangChain.
Even prompt engineering roles that were explicitly open to people with linguistics, philosophy, and other non-technical degrees, typically required familiarity with the more technical side of deploying AI models.
More like this
How to become a prompt engineer
As with any emerging role, there’s no one pathway to becoming a prompt engineer, but there are a few things that can help.
First, good prompt engineers should have an understanding of core concepts like prompt chaining and retrieval-augmented generation (RAG). A strong understanding of how large language models are developed and the transformer architecture they rely on will also go a long way.
Bartholomew recommends The Prompt Engineering Guide as a great place to start. Josh Ferry Woodard, a non-technical prompt engineer at Jamworks, spent months watching YouTube videos and reading papers about AI models because they fascinated him.
Second, prompt engineers need to be analytical, curious, and empathetic. As Ferry Woodard explains, the smallest tweaks to a prompt can make huge differences to the output – and not always in the ways you expect. Every AI model is a black box; if you want tools that respond predictably in every situation, traditional programming languages are a better path.
Finally, you need to be open to opportunities. Prompt engineering is a new and varied role. There are job listings that call for masters degrees in computer science and others that just want someone eager to experiment with AI models. While technical skills are always going to be an advantage, they aren’t necessarily required for every position. Check job boards and if something sounds good, apply.
Prompt engineer salary expectations
According to Glassdoor, the typical prompt engineer salary in the US is between $104,000 and $160,000 based on more than 40 submissions. In the UK it’s a lower range of between £39,000 and £71,000. Based on our reviewing of jobs listings, more technical prompt engineering positions offer better compensation than less technical ones.
While prompt engineering may not be quite the high-paying non-technical position headlines have painted it as, LLMs and AI models are having a significant impact across a wide range of different organizations. Anyone who can deploy and work with AI to get great results is likely to be in a good position going forward – whether you are employed as a prompt engineer or just use prompt engineering in another role.
