Despite impressive achievements, large language models like Chatgpt are still prone to mistakes – and with confidence in a frightening way both correctly and intentionally wrong answers. The problem is the uncertainty that such models work. However, Swiss researchers found a way to reduce this uncertainty, and AI developed a method that allows you to “feed ıyla with the relevant data, so that the answers became more accurate. Work Published on Preprint Portal Arxiv.
The algorithm selects the most complementary data and not only repeats known information. This is obtained by analyzing the vectors of the semantic connection between the words in the multidimensional area. The closer the direction of the vectors, the higher the correlation. However, unlike the classic approach of the “closest neighbor”, Sift selects different, but complementary information.
Example: “How old and how many children does Roger Federer have?” The traditional AI can be “stuck” on the date of birth and may ignore the information about children. And Sift finds both age and family information and supports one aspect to another.
Innovation is not only for quality: less information – less load in the system. The adjective allows you to adapt the calculation amount to the complexity of the problem, which means creating reliable models with lower resources. In the tests, the researchers reached the performance of the best AI, while the models were 40 times less.
Researchers emphasize: Adjective, for example, can be useful for data analysis to determine which laboratory indicators are important for diagnosis and which ones are secondary. This leads to new practices in medicine, science and business.
Previously, scientists saidWe can find evidence of the existence of a foreign life.
What are you thinking?
Source: Gazeta
Barbara Dickson is a seasoned writer for “Social Bites”. She keeps readers informed on the latest news and trends, providing in-depth coverage and analysis on a variety of topics.