10 emerging AI terms that will impress your geekiest health tech colleagues

It used to be enough to drop ChatGPT, Generative AI, or OpenAI to impress your friends and co-workers. But then things started moving very quickly. Saying just AI is almost the equivalent of talking about oxygen or this new thing called a cell phone. AI terminology has become ridiculously specific and complex. But while not yet mainstream the growing generative AI-using population is starting to grasp what were once cryptic terms and acronyms.  

So to assist in your efforts to continually impress and baffle, we offer the following ten terms that are not quite mainstream yet unless you happen to be living in AI geekdom.  

• Prompt Engineering: This term is an icebreaker before you test the others below! Prompt engineering involves crafting specific and clear instructions or questions to guide AI models, enhancing their ability to generate accurate, relevant, and contextually appropriate responses based on the provided input. This is on the verge of becoming mainstream. 
• Light LLMs (Light Large Language Models): This could be considered the GLP-1 of AI coding! These are smaller and less resource-intensive models designed for applications where massive LLMs are impractical. They aim to provide efficient performance while reducing computational overhead. ​ 
• Multistage LLM Chains: This concept involves connecting multiple LLMs to perform a series of tasks sequentially, enhancing the capabilities of AI systems by leveraging specialized models for different stages of a process​.
• User-in-the-loop AI (UITL): A workflow that requires users to be looped into any stage of the AI system development pipeline. Essentially the system will advise the user that will lead them to improve the performance of the system. This as opposed to the user guiding the algorithm. Yes, a bit anthropomorphically scary?
• Hallucination management:  Refers to managing incidents when LLM-generated content is nonsensical, blatantly factually incorrect, or in some cases totally improper like asking the user to leave his spouse. When combined with prompt engineering, output will be increasingly accurate and targeted to the prompt. 
• Diffusion Models: These are generative models that generate new data samples by iteratively adding and removing noise, offering a probabilistic approach to data generation. They are particularly useful in creating high-quality synthetic data​  
• AI Models as a Service (AIMaaS): A service model where AI models, including generative models, are offered for inference and fine-tuning as consumable services by cloud providers. This approach simplifies the deployment and integration of AI into various applications​ 
• Provenance Detectors: These tools are designed to identify whether content (text, audio, or video) was produced using generative AI, helping to manage and verify the authenticity of AI-generated content​. This has become the latest equivalent to the plagiarism checkers used by many universities, but now on a much broader scale given the proliferation of AI-generated content in business that may cross intellectual property lines 
• Synthetic Data Generation: The use of generative AI to create artificial data sets, which is becoming increasingly important for training models in scenarios where real data is scarce, expensive, or subject to privacy constraints. While synthetic data has been around for a while the generative aspects will lead to significant growth in the coming years​  
• Retrieval-Augmented Generation (RAG): This technique combines a retrieval system with a generative model to produce more accurate and contextually grounded outputs by using relevant information retrieved from databases or the web during the generation process​. This is a morph of Prompt Engineering, Hallucination Management and Provenance.  

OK, you’re now armed with some of the trendiest AI terms for your next corporate cocktail party or Zoom call. But given the rapidity of generative AI growth, these will become “your father’s AI terms” in short order, so follow DHI for regular updates!