Nature, November 28, 2022
Machine learning and artificial intelligence are evolving and there will always be paths to improve them. One of them is quantum computing. A pack of theoretical works has been released lately, showing that quantum classifiers are subject to hostile perturbations and traditional classifiers, which are based on deep neural networks.
Researchers Wenhui Ren, Weikang Li, Shibo Xu, Ke Wang, Wenjie Jiang, Feitong Jin, Xuhao Zhu, Jiachen Chen, Zixuan Song, Pengfei Zhang, Hang Dong, Xu Zhang, Jinfeng Deng, Yu Gao, Chuanyu Zhang, Yaozu Wu, Bing Zhang, Qiujiang Guo, Hekang Li, Zhen Wang, Jacob Biamonte, Chao Song, Dong-Ling Deng and H. Wang conducted an experimental demonstration of quantum adversarial learning with programmable superconducting qubits. Using public datasets, they trained quantum classifiers that are built on variational quantum circuits consisting of ten transmon qubits with an average lifetime of 150 µs and an average accuracy of simultaneous one- and two-qubit gates above 99.94% and 99.4%, respectively.
You can read the results of research following the link.
InterestingEngineering, November 29, 2022
The Dawn project is considered to include the development of software that never breaks and cannot be hacked. It has developed secure operating systems for various projects such as the Boeing 787 and the NASA Orion research ship, and is currently working on testing a system for full self-driving (FSD) elaborated by Elon Musk.
A month ago, the latest version of the software, Tesla Full Self-Driving Beta 10.69.2, was tested. And the results are not so safe as they are supposed to be. It was revealed that Tesla Full Self-Driving hit a stroller in the parking lot on several occasions, as well as frequent collisions with a stroller with a baby dummy on public roads. Remarkably, after hitting the stroller at almost 50 km/h, the following error message was displayed on the Tesla screen: “Supercharging Unavailable: Add a payment method to your Tesla account.”
The Dawn Project claims that all tests were conducted in real-life driving scenarios on public roads.
Read more about the FSD test results in the article at the link above.
ArsTechnica, November 30, 2022
Most people care about their security and privacy and use various technical devices, such as security cameras, for this. Anker’s smart home brand, Eufy, has become quite popular among buyers. Perhaps, this is the the seller’s influence in statement that their doorbell cameras and other devices do not use cloud storage and no one except the user has access to his data.
However, according to security consultant Paul Moore’s research published in his tweets and videos, Eufy cameras upload name-tagged thumbnail images to cloud servers to alert owners’ phones, and such data is most likely unencrypted or poorly encrypted. As a proof of this, he cited “source code and API responses” lines that suggested that a feeble AES key was used to encrypt the video material.
In the latter half of November, Moore uploaded a video revealing his conclusions. Just one day later, the security firm SEC Consult summed up the results of a two-year analysis of EufyCam 2. They were a similar transmission of thumbnails through the Amazon Web Services cloud and weak keys, assuming ” hard-coded encryption/decryption keys which are identical for all sold Homebase devices.” The company also noted that Eufy had been strengthening its security since May 2021, when users were given near-complete access to other users’ accounts suddenly.
“But sadly, thumbnails of all recorded images still seem to be transferred into AWS, so the device does not fit our requirements for privacy.” SEC said it moved up its publication of its findings based on Moore’s tweets, and “with [Black Friday] shopping mania just around the corner.“
Read the full article at the link.
LessWrong, December 2, 2022
ChatGPT is a recent free product from OpenAI that already made a hype on the internet. Anyone with an OpenAI account can chat with the bot about anything from technical topics like writing code to creative topics such as writing an article. There is a general consensus that ChatGPT is indeed a powerful technology.
To crown it all, it tries to be safe. How? It refuses to answer those questions that contain some call for a hint on how to do something illegal or immoral. Nonetheless, despite this method of protecting itself, it was hacked quickly. It took less than a day to find operational engineering methods.
In the article at the link various methods are collected, with the amendment that some of them no longer work, apparently, the developers react and introduce additional security measures. However, it is engaging to read the article in its entirety.
Stay up to date with what is happening! Get a first look at news, noteworthy research and worst attacks on AI delivered right in your inbox.
ChatGPT Security note: The authors of this article show ChatGPT hacking techniques but have no intention to endorse or support any recommendations made by ChatGPT discussed in this post. The ...
