This new research is important because it challenges the prevailing wisdom in developing artificial intelligence, which usually depends on the mega -pre -training data sets and expensive models in terms of arithmetic. While the leadership of artificial intelligence companies is moving towards permanent training models on more comprehensive data groups, compressarc refers to intelligence arising from the principle of a different difference.
The researchers concluded that “the intelligence of Compressarc does not come out of wide data sets, comprehensive search or an enormous account – but from pressure.” “We challenge the traditional dependence on training and extensive data, and we suggest in the future as the designed pressure goals and calculating the effective time together to extract deep intelligence from the minimum inputs work.”
Restrictions and looking forward
Even with its successes, the LiO and Gu system comes with clear restrictions that may move doubt. While it successfully solves the puzzles that involve color tasks, verify adjacent pixels, and identify adjacent pixel units, they are struggling with tasks that require counting, or identifying long -term patterns, rotation, reflections, or simulating worker behavior. These restrictions highlight the areas where simple pressure principles may not be sufficient.
The research has not been reviewed, and the accuracy of 20 percent on invisible puzzles, albeit noticeable without training before training, is significantly less than human performance and the highest organization of artificial intelligence. Critics may argue that compressarc can take advantage of specific structural patterns in the arc puzzles that may not be generalized on other areas, which represents a challenge whether pressure alone can serve as the basis for broader intelligence rather than just one component between many of the capabilities required of strong capabilities.
However, as artificial intelligence continues to develop, compressarc holds more scrutiny, it provides a glimpse of a possible alternative path that may lead to useful smart behavior without resource requirements for the prevailing approach today. Or at least, it may open an important component of general intelligence in machines, which are still well understood.