DIGITALLY SECURE: NEW CIPHER TO PROTECT COMPUTERS FROM CACHE ATTACK
Modern computer systems are increasingly vulnerable to side-channel attacks from cache, a computational component that stores data for previously asked-for data to be accessed again faster subsequently. It could be the result of an earlier computation or a copy of the data served elsewhere. However, cache side-channel attacks allow sensitive information like secret keys and passwords to be stolen through extraction. Although cache randomization is known to be an effective measure, identifying a secure mathematical function to support it has remained a challenge.
Now, a team of international researchers has developed a highly efficient cipher system for cache randomization. A cipher is an algorithm that performs encryption or decryption, which are steps that can be followed as a procedure. The new cipher is acknowledged as innovative as it addresses cache side-channel attacks to provide enhanced security with better performance. A side-channel attack uses shared cache in the central processing unit (CPU) of the same device and compromises the privacy of the system, besides jeopardizing the system’s security by leaking information through channels which are not intended to transmit it. The new cipher, called SCARF, is developed through a mathematical formulation and based on modelling of side-channel attacks, and has been found to offer robust security.
Its performance has been found to be excellent while completing the cache randomization process. The performance and the process of SCARF were validated by system-level simulations and hardware evaluations before being convinced of its robustness and security-enhancing capabilities, which open up avenues towards safeguarding not just individual computers, but also ensure a more secure information society in the future. The team that developed SCARF comprised members from Tohoku University, CASA at Ruhr University Bochum, and NTT Social Informatics Laboratories at Nippon Telegraph and Telephone Corporation.
Dance of AI
BIG CONCERN: GPT-3 BETTER THAN HUMAN STUDENTS?
A new finding has fuelled fresh concerns about the way artificial intelligence is growing, and whether it would threaten humans in future. A research team at University of California Los Angeles (UCLA) published in Nature Human Behaviour has shown that the artificial intelligence language model GPT-3, when put to intelligence tests comprising reasoning problems and standardised tests, performed just as well as undergraduate students who were given the same kind of tests. Having found that, the research team has raised questions on whether GPT-3 has learnt how to mimic human reasoning, and whether it was a new cognitive process, or just a byproduct of its huge language learning dataset.
The problem that faces the UCLA researchers, as much as anyone else concerned with the rise in reasoning skills supported by artificial intelligence, is that they are unsure how the reasoning abilities of GPT-3 work without access to its inner workings that are guarded by OpenAI, which is the company that created it. However, what has foxed them is that while GPT-3 excels in some tasks compared to humans, it still falls short while attempting some others. But that does not lessen the concerns considering the flashes of excellence demonstrated by GPT-3, threatening to overshadow human capabilities.
The team put GPT-3 through a test that asked to predict the next image through a maze of shapes placed in a complicated manner. They raised the challenge by converting the images into a text format for GPT-3 to “see” the images, which ensured that the AI language model was not familiar with the questions prior to the test. GPT-3 scored 80% compared to 60% by the UG students, but also made similar mistakes as the students, which raised suspicions about GPT-3’s cognitive performance being on the same lines as the human mind. Moreover, even in the standardized tests, GPT-3 performed better than the average score of the students.