In many ways, 2023 marked the dawn of a new era in AI. It was the year that people everywhere began to realize the true potential and power of AI technology. From viral chatbots to government acknowledgment of AI risks, 2023 was a turning point for artificial intelligence.
But beyond these developments were several groundbreaking innovations that set the stage for a new wave of AI capabilities. Here are three of the most significant innovations from the past year:
Multimodality
The buzzword “multimodality” may sound complex, but its implications are profound. It refers to the newfound ability of AI systems to process a myriad of data types, including text, images, video, audio, and more.
One of the most notable breakthroughs this year was the introduction of powerful multimodal AI models. OpenAI’s GPT-4, for instance, ushered in an era where users could input images in addition to textual data, allowing the AI to “see” and process the contents of images. This opens the door to a wide range of possibilities, such as seeking dinner ideas based on a photo of your refrigerator’s contents. Moreover, OpenAI’s ChatGPT now supports voice interactions, expanding its capabilities even further.
In a similar vein, Google DeepMind unveiled its Gemini model, which also boasts image and audio processing capabilities. This was demonstrated in a video where Gemini accurately identified a duck based on a simple drawing and generated an image of an octopus plushie in response to a question about yarn. The future of AI, according to Google DeepMind co-founder Shane Legg, lies in the transition to fully multimodal systems, opening up a wealth of possibilities in video processing and beyond.
It’s not just about making models more useful; it’s about unleashing their full potential by training them on vast datasets of images, video, and audio, which provide a deeper understanding of the world than text alone. The consensus among top AI companies is that this new training data will empower models to become even more capable and powerful.
