Daniel Restrepo, MD, a physician at Massachusetts General Hospital, conducted two studies comparing the diagnostic reasoning abilities of large language models (LLMs) to human physicians. Published in *JAMA Internal Medicine* and the *Journal of Hospital Medicine*, the research explored whether AI could reduce clinical reasoning errors that often lead to misdiagnoses. The first study compared an LLM and a human doctor in diagnosing a real medical case, revealing that the AI was slower to integrate new data points but arrived at the correct diagnosis. The second study assessed GPT-4 against resident and attending physicians, noting the AI's comparable reasoning abilities but also its tendency toward verbosity and higher instances of incorrect reasoning. The findings suggest LLMs could potentially augment, but not replace, human clinicians, highlighting the need for further study and careful implementation to address biases and data safety.

Research Spotlight: Head-to-Head Comparisons of Generative Artificial Intelligence and Internal Medicine Physicians Mass General Brigham

The rapid advancement of the digital afterlife industry is bringing the possibility of interacting with virtual reconstructions of deceased loved ones through AI, VR, and other technologies. Companies in this niche market, such as HereAfter and MyWishes, are developing digital personas based on the data left behind by individuals, raising significant emotional and ethical questions. These technologies offer both comfort and potential psychological harm, blurring the lines between reality and simulation. Issues around consent, privacy, and the potential for misuse call for updated legal frameworks and ethical guidelines, including informed consent and data security measures, to ensure these digital interactions honor the deceased while supporting the emotional health of the living.

An eerie ‘digital afterlife’ is no longer science fiction. So how do we navigate the risks? theconversation.com

The article from The Wall Street Journal discusses how scammers are increasingly using artificial intelligence (AI) to outsmart both consumers and financial institutions. Techniques include mimicking voices for fraud calls and creating realistic phishing emails. The advanced AI tactics present significant challenges for cybersecurity, requiring new approaches to protect sensitive information.

AI Is Helping Scammers Outsmart You—and Your Bank — The Wall Street Journal Wall Street Journal

Researchers at cybersecurity firm Qualys have identified a critical vulnerability in the widely used OpenSSH secure communications protocol, named "regreSSHion". This flaw, affecting nearly 14 million instances, could allow attackers to gain full access to systems and bypass firewalls, though it is challenging to exploit under typical conditions. The vulnerability, CVE-2024-6387, was re-introduced in 2020 after being fixed nearly a decade ago. While experts caution against overhyping the severity, they emphasize the importance of zero trust and mitigating risks. The bug primarily impacts older, 32-bit Linux systems and highlights the need for continued efforts toward using memory-safe languages to secure open-source ecosystems.

Researchers uncover rare, difficult-to-exploit OpenSSH vulnerability cyberscoop