July 13, 2023

Harnessing the Power of Generative AI

Harnessing the Power of Generative AI

Source of image: Generated by Dall-e 2 with the prompt: "A photo of a hamster on a skateboard in Time Square"

The Power of Generative AI

Generative AI models leverage advanced machine learning algorithms to predict the next sequence based on previous sequences, opening up a world of possibilities for innovation and automation. The technology, capable of creating a wide array of content from text to images and even code, is revolutionizing industries from marketing to software development, entertainment, and healthcare.


Generative AI is not only becoming increasingly accessible and cost-effective but also offers a level of complexity and speed that far exceeds human capacity. Research by McKinsey suggests that up to a fifth of current sales-team functions could be automated with the rise of Generative AI. As an example, the technology can be transformative in customer experience with the provision of hyper-personalized content based on individual customer behavior, persona, and purchase history.

Software Development

Generative AI, particularly Large Language Models (LLMs) like GPT-3, has proven its versatility and can be applied to various tasks beyond generating text. One capability that has been especially interesting for software developers is generating computer program code. LLMs can assist developers by predicting the next lines of code based on the code already written or based on a prompt that describes the functionality of a certain piece of software. This can significantly accelerate the software development process.

Gaming and Entertainment

Imagine a world where video games offer endless narratives and adventures, where each playthrough is a unique experience. This is no longer a distant dream, but a reality being shaped by Generative AI. The concept, which has already been applied by Netflix in the movie Black Mirror, allows viewers to choose their own adventure, leading to different outcomes for the characters. This same principle can be applied to gaming, creating an infinite array of narratives and adventures that make games more replayable and captivating than ever before.

Another exciting application of Generative AI in gaming is the creation of realistic non-player characters (rNPCs). These AI-generated characters can have unique personalities, appearances, voices, body movements, goals, and memories, making games more engaging and immersive. Imagine interacting with an rNPC that remembers past encounters and reacts differently based on your previous actions.


According to a research webinar by CB Insights, there are three key areas of the healthcare sector where generative AI is booming the most;

  • care delivery and navigation, where generative AI tools are being used as "AI co-pilots for doctors," helping to automate electronic health record (EHR) workflows for physicians alleviating clinical burdens through medical conversation summaries;
  • digital therapeutics and wellness tools, for example Replika, a chatbot designed to provide users with emotional support. However, there are significant risks associated with using large language models in mental and emotional health services, as evidenced by some incidents where the chatbot encouraged dangerous behaviors.
  • drug research and development, predicting the properties of novel proteins and drugs, optimizing new candidates for drugs more quickly.

Computer Vision and Data Labeling

We, at SUPA, too are experimenting with the use of Generative AI in our space. Most recently, with the Segment Anything Model by Meta (SAM). SAM is capable of generating segmentation masks for images it has never seen before, based solely on a text prompt. This is a significant advancement in the field of computer vision, as it allows for more flexible and adaptable image segmentation. We believe that this technology could significantly change how we work, making our data labeling processes more efficient and accurate. 


As we move forward, the safe and reliable integration of Generative AI into various sectors will be crucial for realizing its full potential. It has the power to automate tedious and error-prone operational work, such as bringing years of clinical data to a clinician's fingertips in seconds, thereby revolutionizing industries.

However, the power of Generative AI is not without its challenges. The unpredictability of the AI models and the resource-intensive nature of the technology can pose significant hurdles. This is where high-quality, human-annotated data comes into play, making this powerful technology more accessible to a wider range of businesses.

A prime example of this in our space is the utilization of human data labeling to improve the Segment Anything Model (SAM) by Meta. The development of SAM introduced a new dataset, SA-1B, which includes over 1 billion masks on 11 million licensed and privacy-respecting images. This dataset was created through extensive data labeling, highlighting the crucial role of this process in training generative AI models. Data labeling allows these models to understand the correlation between different elements in an image, enabling them to generate accurate segmentation masks.