Bill Gates once stated, "Most people overestimate what they can achieve in a year and underestimate what they can achieve in ten years” (Alammar, 2023: Online). The same can be inferred about people’s expectations of innovative technologies. These expectations, in most cases, are unrealistic and far-fetched. The last time the tech industry was seduced by a deep learning frenzy, we were promised self-driving cars by 2020. Seeing as in 2023 we are not going to work in one of them, this points to an unrealistic reality.
However, Generative AI is nonetheless an exciting and promising field and should not be discarded just because it is not easily understood, implemented, or achieved. After all, nothing substantial is ever achieved quickly.
Generative AI can be understood as “a set of algorithms, capable of generating seemingly new, realistic content—such as text, images, or audio—from the training data.” The most powerful Generative AI algorithms are built on the foundation models that are trained on a vast quantity of unlabeled data in a self-supervised way to identify underlying patterns for a wide range of tasks (Boston Consulting Group, 2023: Online).
ChatGPT – Chat “Generative Pre-trained Transformer” (GPT) is an AI large language model (LLM) which deploys deep learning and autoregressive (AR) modelling. AR modelling is simply a model that predicts a future outcome in a series, or body of text based, on previously observed outcomes of that sequence.
The success and failure of Generative AI depend on your point of view, more specifically, on the cherry-picked demos and reliable use cases used to prove the success of Generative AI. However, some groups have disregarded results from Generative AI because, “the average rate of getting correct answers from ChatGPT is too low” (Alammar, 2023: Online). This is an example of a use case where people expected the model to reliably generate an exact resolution for a complex set of problems.
However, there are other use cases (and workflows) where these models are capable of more reliable results. Among them are neural search, auto categorisation of text (classification), copywriting suggestions and brainstorming workflows for generation models.
Impressive demos will continuously be introduced, and they are part of a community discovery process to evaluate the limits and new possibilities of these technologies. As with any recent technology, keep questioning the usefulness of the cherry-picked examples, recognise their uncertain timelines, and invest in the robustness and reliability of AI systems and models.
Once we think of Generative AI (model/tool) as a component, we can start to compose more advanced systems that use multiple Generative AI models or tools. If we make each of the Generative AIs address a specific area, we can combine multiple Generative AIs to address a larger environment. Another way of thinking of this is to look at the motor vehicle; it is not one thing but how all the individual parts work together to form a tool to transport people.
“Generative AI is only possible because larger, better models trained on massive datasets enable AI models to make better numeric representation of text and images” (Alammar, 2023: Online). For creators of Generative AI, it is important to know this enables a wide variety of possibilities, such as neural search. Neural search is a search system that uses language models to improve on a simple keyword search and enables searching by meaning.
There are many promising Generative AI and AI developments, but they must be assessed and aligned with your business needs.
Sources:
Alammar, J. 2023. What's the big deal with Generative AI? Is it the future or the present? [Online]. Available at: https://txt.cohere.ai/generative-ai-future-or-present/ Accessed: 2023.04.02
Boston Consulting Group. 2023. Generative AI. [Online]. Available at: https://www.bcg.com/x/artificial-intelligence/generative-ai Accessed: 2023.04.02