Tim DaviesCEO Projekt Rising - The expert AI partner for SMEs

Future developments in AI technology

10 Dec 2024

Watch on YouTube

Single-speaker YouTube monologue (Tim Davies) forecasting near-term AI developments: rapid increases in LLM capabilities, emergence of robust multimodal understanding (audio, video, images, text) with models responding natively in those formats, and parallel development of alternative model structures using joint embeddings rather than exclusively transformer architectures (e.g., training on altered video pairs to learn differences).

The developments over the next year are going to be fascinating. We're going to see ever increasing capabilities in terms of the intelligence and ability to answer questions within the large language models. The ability to understand content regardless of its modalities. So whether it's audio, video, image, words won't matter. And the ability for the models to understand them and natively respond in those same formats will be coming quickly. But we're also seeing developments of similar technologies that are based on different structures. So instead of being based on the transformers, like a lot of these large language models, all of the large language models are looking at the ability to understand something based on joint embeddings. So potentially taking a video, making a copy of that video, blocking off part of it, and then feeding millions of these videos into a model to have the model understand the differences between the two videos.