MIT Technology Review analyzes next trends in LLMs and shifts in AI research

The 'transformer,' the heart of AI, has reached its limits. MIT Technology Review analyzed that the 'transformer,' a core technology in artificial intelligence (AI), is suffering from efficiency issues where data processing costs surge as the model grows. Since Google introduced it nine years ago, the transformer has served as the core engine for all major Large Language Models (LLMs).

However, as the amount of text to be processed increases, the 'attention' function, which focuses on important information, is proving inefficient, and limitations in processing large amounts of information simultaneously are emerging.

Four new technical alternatives have been proposed to resolve the transformer's bottleneck. These technologies aim to improve the model's processing speed and efficiency.