There is an intense, ongoing competition to develop new AI models with increasingly massive numbers of parameters, resulting in dramatic improvements in AI's ability to recognize and utilize language and images.
A new kind of AI is emerging that will be able to learn and solve problems by itself and be applied to a multiplicity of uses not limited to specific fields.
While services using AI such as smart speakers and automated inspection tasks have become the norm, there are still challenges with AI, such as underperforming capabilities and the need for manual collection of learning data. To solve this problem, AI will not only massively increase the numbers of parameters, but also become multimodal and self-learning.
Natural Language Processing in AI continues to improve its capabilities by increasing the number of parameters and learning large amounts of data. This massive AI is enabling the creation of new AI software with advanced capabilities in applications such as translation, summarization, image and voice.
In addition, AI is gaining new capabilities using multimodal data. Traditionally, there have been separate AI systems for different types of data, such as language or images. Multimodal AI combines these multiple data types, for example, to generate images that match verbal commands or process any type of data.
AI is also evolving to overcome the amount of human time and effort required to collect learning data. New data collection methods, such as self-supervised learning and synthetic data, are addressing this need. As technology evolves, the IT world continues to find new ways to improve and exploit AI’s capabilities. Getting this right will make utilization of AI successful.