For a long time, technology giants hope that AI systems are developed because they will enter the strong artificial intelligence (general artificial intelligence – AGI), which means that they can be comparable or outstanding with humans, by simply expanding. However, when the performance of the final models on the plateau, optimism was replaced by skepticism.
About 76% of AI experts consider the difficult achievements of Muslims or it is difficult to be able to maintain current methods.
Survey results of 475 representatives of competent industry have been included report Associations on the development of artificial intelligence (the progressive association of artificial intelligence) – an international scientific society based in Washington.
This is a noticeable change in the mood compared to the optimism of the scale that is all necessary, inspired by technology companies from the beginning of the article of the article by the generation by 2022. The impressive indicators in recent years have been achieved mainly by the transformer architectural models, which has increased their efficiency as they studied larger and larger data. However, in the latest releases, their process seems to stop, showing only small improvements in quality.
Large investments in expanding scale, not accompanied by some equivalent efforts to understand what is happening, seems to always not suitable for me. I thought that about a year ago, people became clear that the advantages of expanding in the sense were often accepted to become meaningless, Professor Stuart Russell from the University of California to Berkeley, a member of the Commission, made a report.
However, in order to support their ambitions in the fields in the next few years, technology giants plan to spend a total of trillions of dollars for data centers and chips.
Whoever claims to focus on people from their work is also very high, 80% of scientists are surveyed.
Systems are claimed to be corresponding to human abilities – for example, in mathematical or programming tasks – still make stupid mistakes. They can be very useful as auxiliary tools in research and encryption-but they will never replace everyone's employees, Mr. Thomas Dertterich from Oregon University, who is also involved in the reporting preparation.
Now, AI developers have focused on expanding the size of the output time, implying the use of more electrical models and processing longer requirements before the answer – but this is not capable of becoming a cure for AGI, ArVind Narayan from Princeton University.
Technology giants strive to create AGI, but this definition is still vague: Some people believe that this is super night is the superiority of artificial intelligence for life, others implying the presence of a body to interact with the environment, others believe that AI should bring at least $ 100 billion.