Research Highlight: LLM Can Handle More Text Than We Imagine

2023-06-19

AI21 Labs' research team is responsible for developing artificial intelligence platforms such as Human or Not, Wordtune, and Jurassic 2 for generating text. The team has discovered a new method to overcome the challenges faced by large-scale language models (LLMs), which previously had high costs and were impractical when processing large amounts of text. These research findings come from a study where researchers found that by making two simple changes to the attention mechanism, LLMs can read multiple texts simultaneously, thus avoiding the initial cost problem. Through extensive testing, the team has demonstrated that these models have built-in parallel reading capabilities, making the processing of multiple texts more efficient and accurate. Let's say you own a hotel and want to classify customer reviews based on various parameters such as cleanliness, check-in process, and facilities. Previously, LLMs may have encountered difficulties in scanning and categorizing reviews comprehensively. However, by allowing LLMs to scan texts simultaneously at different intervals, they can improve their ability to classify both existing and future reviews. Source: https://insidebigdata.com/2023/06/16/research-highlights-llms-can-process-a-lot-more-text-than-we-thought/

RECENT AI NEWS

RECENT AI TOOLS