Former "Hint Engineer", Now "AI Engineer"

2023-07-05

Prompt engineering is actually becoming a full-time job that involves not only prompting ChatGPT but also full-stack development. The boom of generative AI has brought many new opportunities. In the past year, the number of positions for generative AI engineers has almost doubled. This creates a new abstraction layer and specialization for developers, a complete dimension shift. Prompt engineers are often referred to as people who provide prompts to ChatGPT or similar software. However, this is actually becoming a full-time job. In a recent popular blog post, someone suggested that instead of calling new developers "prompt engineers," we should call them "AI engineers." However, this is not to be confused with machine learning engineers who handle the heavy workload of building models from scratch. Compared to these machine learning or language model engineers, we will definitely see the rise of more AI engineers. Andrej Karpathy also expressed a similar view on this issue. He said that the term "prompt engineer" can be misleading and even awkward because this role requires much more than just prompting. The duality of AI engineers The role of AI engineers is expected to evolve, which can be both good and bad. Many experts in the field believe that with the improvement of language models, there may not be a need for what needs to be done currently after generating code from these tools. The illusion is only temporary. On the other hand, as the field develops, more specialized knowledge is also required. This is true in every field. There will be AI engineers, and then there will be full-stack prompt engineers. Just as there are disciplines like DevOps engineers, analytics engineers, and data engineers, there will be roles that handle different aspects of AI tasks. In the past, traditional machine learning typically involved finding data, building models, and then launching products. Now, with AI engineers, the process has reversed as they enter the industry after trying out the product. They build a product using tools like ChatGPT's API, called Jasper or any other GPT-based tool. Then, they proceed to fine-tune and expand the model using data. The latter two steps require more technical knowledge than just prompt engineering. Hence, the rise of AI engineers. With the availability of APIs, AI engineers can transition to more application-based roles or delve deeper into foundational knowledge. This can be seen as a shift from MLops to LLMops. Considering the nature of the work, these AI engineers need to understand every step of the LLM application stack. This requires them to expand their knowledge beyond prompt engineering. So, while becoming a developer has become easier, the challenge of becoming the best in this field is vast and extensive. Is this Software 3.0? Often, companies need people who know how to use and launch products or tools rather than those who know how to build them. For example, a user on HackerNews argued, "If we have React engineers, why can't we have AI engineers?" This is true; it's just a demand for different skills. LLM/AI engineers also need a systematic experimental and observability workflow, especially in prompt creation and system development. Challenges like illusions and reasoning gaps are common in LLM, leading to a consistent understanding that these agents are unreliable. Both traditional models and LLM systems require some adjustments for optimal performance. Traditional models can be improved through training and hyperparameter selection, while LLM systems can be fine-tuned by modifying prompts and linking LLMs. Now, this can be done not only by LLM researchers but also by prompt engineers. In the past, software development relied on manually writing instructions in programming languages. Machine learning and neural networks enable systems to learn patterns and make predictions based on training data. This is considered Software 2.0, where models are trained to perform specific tasks. Microsoft claims that everyone is a developer. Karpathy says that our most popular programming language is our own English. Software 1.0 was classical coding, transitioning to machine learning and neural networks, i.e., Software 2.0. The next transformation for prompt-based developers should be Software 3.0. However, Karpathy states that we are still at the second level of abstraction because "we are still prompting on human-designed code, just in English," which is still a product of Software 2.0. Source: https://analyticsindiamag.com/prompt-engineers-then-ai-engineers-now/

RECENT AI NEWS

RECENT AI TOOLS