Plus, Dubai rolls out Robocop 1.0
OpenAI’s GPT-4 shows promise, but it faces vulnerabilities and biases, reveals Microsoft-backed research
OpenAI’s GPT-4, a language model, has been found to be more trustworthy than its predecessor, GPT-3.5, according to a study supported by Microsoft. However, researchers discovered that GPT-4 is susceptible to jailbreaking and biases. While the model is generally better at protecting private information and avoiding biased results, it can also be manipulated to ignore security measures and leak personal data. Researchers emphasized that these vulnerabilities have not been identified in consumer-facing GPT-4-based products due to mitigation measures. The study’s benchmark results have been shared to promote collaboration and further development of trustworthy AI models.
Convicted Fugees rapper Pras Michel’s lawyer used AI to draft bungled closing argument
Convicted Fugees rapper Prakazrel “Pras” Michel’s lawyer may have jeopardized his client’s defense by using an experimental generative AI program to draft his closing argument. Michel’s new counsel from ArentFox Schiff has called for a retrial on the grounds that the AI-generated argument was flawed and failed to properly address critical weaknesses in the government’s case. Furthermore, it has been alleged that Michel’s defense lawyers had an undisclosed financial interest in the company behind the AI program, which compromised their representation of Michel. A motion for retrial has been filed, citing not only the use of the AI program but also the trial judge’s handling of certain aspects of the case as factors that tainted the proceedings. The defense argues that the trial judge allowed the jury to hear information that prejudiced Michel and that the testimony from the lead case agent was improper. Additionally, Michel’s counsel accuses his former defense lawyer of outsourcing trial preparation to inexperienced contract attorneys, leading to inadequate cross-examination of government witnesses.
Leading AI Researcher Warns of Urgent Threats and Calls for Regulation
Yoshua Bengio, one of the leading researchers in AI, has revised his estimate for when AI could reach human-level cognitive competence. He now believes that it could be achieved within a few years or decades, posing significant threats to democracy, national security, and the collective future. In an interview, Bengio discusses his realization of the risks, the taboo nature of discussing AI threats in the research community, and the need for regulation to mitigate these risks. Bengio emphasizes the importance of national and international regulation, as well as monitoring and tracking hardware purchases, to contain the potential harms of AI.
Will AI Replace Architects?
AI has the potential to automate routine tasks in architecture and engineering, allowing professionals to focus on more creative and strategic aspects of their work. While AI is not expected to replace architects entirely, it will bring about new tasks and applications that can complement and enhance their work. From generating design alternatives to analyzing site data, recognizing patterns, coding, and aiding in sustainability efforts, AI is poised to revolutionize the architectural field. Architects are urged to embrace AI and continue updating their skills to remain competitive in a rapidly evolving profession.
The Rise of AI as a Band-Aid for Tech Industry Design Failures
Ed Zitron discusses how technology companies like Apple and Google are using AI as a patch to cover up their poor design choices and lack of user-friendly interfaces. The author argues that instead of focusing on creating better products, these companies are charging users more for AI solutions that fix their mistakes. The article explores how AI-powered features like search engines, voice assistants, and user interfaces not only fail to address the underlying design problems but also give tech giants more control over users’ decisions and experiences.
Dubai Police Introduce AI-Equipped, Self-Driving Patrol Cars to Enhance Residential Security
Dubai Police plans to revolutionize security and safety in residential areas by deploying fully electric, self-driving patrol cars equipped with AI. These high-tech patrol cars feature 360-degree cameras, facial recognition technology, and the ability to detect criminal behavior and read car license plates. The vehicles also have direct communication capabilities with the police’s Command and Control Center and are equipped with an onboard drone. Expected to hit the streets next year, these autonomous police patrol cars aim to boost security coverage in residential zones.
US Restricts Semiconductor Exports to China, Further Escalating the AI Tech Battle
The Biden administration has tightened export controls on semiconductors, limiting the types of chips that can be sold to China. The new rules aim to close loopholes in existing regulations and increase the effectiveness of controls, preventing China from procuring advanced artificial intelligence chips. The measures also expand export curbs to include 21 other countries. While the restrictions target China’s military advancement, they carve out chips used in phones, video games, and electric vehicles. Beijing has criticized the move, accusing the US of politicizing trade and tech issues. US chipmakers saw a decline in stock prices, but companies like Nvidia do not anticipate a significant impact on financial results.