During recent conversations with airline leaders about the GenAI breakthrough, the first question was: what do we need to know about GenAI? My response – not AI generated – reads like this.

GenAI is a recent evolution in the field of Artificial Intelligence (AI):

• In June 2017, a Google team proposed a new network architecture based on attention mechanisms, the Transformer.
• In June 2018, OpenAI published their results ”improving language understanding” using Generative Pre-trained Transformer (GPT) models.
• In November 2022, OpenAI released ChatGPT to the public who discovered the power of the new AI models designed to understand and generate natural language – forming a new category of AI called “GenAI”.

Artificial Intelligence

The field of AI, boosted by computing power, is developing at exponential speed and impacting the world of software and beyond.

AI is software category that grew since the 1990s. It is based on the training of models (neural networks). Training a model means optimizing the parameters of the model based on a training datasets. For example, a model can learn to recognize pictures of handwritten numbers.

The trained model excel at specific tasks. Applications include: radiology, self-driving cars, credit risk management, shopping engines, visual effects, signal processing, predictive maintenance… An AI model may detect a fraud based on historical credit risk patterns. Another AI may make shopping recommendations based on previous customer behaviors.

Generative AI

New models, developed since 2014 and able to generate content – text, image, sound – are called “Generative” as opposed to the analytic, predictive or discriminative models designed for specific task like forecasting, recognition or classification.

The models that are designed to understand and create natural language, are trained on very large datasets of words and called Large Language Models. The size of a dataset is 100 billion tokens (a token can be a word) and the size of the model is 500 billion parameters (a parameter can be a weight).

A LLM takes a “prompt” as an input and responds in natural language. ChatGPT, a conversational tool developed on GPT (OpenAI’s LLM), can handle a discussion similar to one with a human being, with some limitations.

Diffusion models are designed to generate images and sounds. While LLMs can generate texts that sound like humans, diffusion models create videos that look real to most humans.

Future of AI

In the future, we can expect GenAI to integrate with business processes and systems, instead of using prompts. GenAI models will be natively embedded in the future generation of business systems.

In relation to the physical world, AI and GenAI will integrate with robots to create “humanoids”, i.e. robots able to make sense of the physical world and to perform physical tasks as human beings would do.

Eventually, new breakthroughs (models, tools…) will take AI beyond natural language and videos towards other human capabilities, e.g. planning, reasoning, feeling.

Conclusion

GenAI is a breakthrough in AI around natural language and more. More innovation is coming with the evolution and integration of AI. Now is the time to rethink your business processes and systems. Contact Threedot to support you with your AI projects.