OpenAI has introduced GPT-4o, an enhanced version of its AI model, coined 'omni' for its ability to process text, speech, and video. It improves upon the previous GPT-4 Turbo by incorporating speech alongside image and text analysis, enabling more dynamic and responsive interactions across various applications, including the updated ChatGPT chatbot. GPT-4o will gradually be integrated into OpenAI’s products, offering improved multilingual capabilities, faster performance, and will initially be available to a select group of partners to manage risks associated with audio misuse. Additionally, updates to the user interface and expansion of capabilities like memory and file management are set to enhance user experience and accessibility.

OpenAI debuts GPT-4o 'omni' model now powering ChatGPT | TechCrunch publication

OpenAI has unveiled a new multimodal version of its desktop GPT, dubbed GPT-4o, which enhances user interaction through text, voice, and visual prompts, boasting nearly human-like response speeds. This upgraded model can interpret and respond to diverse inputs like photos, documents, facial expressions, and handwritten information swiftly and effectively. Although these capabilities mark significant progress, analysts note that OpenAI is catching up to competitors like Google, who previously introduced advanced multimodal models. Additionally, GPT-4o is set to include improved memory functions, enhancing its ability to learn from previous interactions. Despite its advancements, GPT-4o's context window size remains the same, and it faces challenges regarding potential misuse in real-time applications.

OpenAI announces new multimodal desktop GPT with new voice and vision capabilities publication

The article "The Power of Motherhood: Supporting Maternal Talent in the Workforce" by Sarah Richardson and Isala Gray discusses the professional challenges faced by mothers due to biases and systemic barriers. It highlights the motherhood penalty, which results in reduced pay and perceived competence, and points out the need for employers to support maternal talent via inclusive policies, flexible work arrangements, and considering the dual demands on working mothers. The authors suggest ways employers can help, such as implementing family-friendly benefits and fostering an environment that values caregiving openly. The piece advocates for a shift towards recognizing and leveraging the skills of mothers in the workplace to foster an equitable and supportive work culture.

The Power of Motherhood: Supporting Maternal Talent in the Workforce publication

Researchers have uncovered a critical vulnerability in almost all virtual private network (VPN) applications called TunnelVision, which compromises the core function of VPNs by routing some or all traffic outside of the encrypted tunnel, potentially exposing user data to interception or alteration. This attack exploits the DHCP option 121 to reroute traffic through an attacker-controlled server instead of the VPN's encrypted tunnel, affecting users on all operating systems except for Android, which is not susceptible due to its lack of implementation of option 121. Despite various mitigation strategies suggested, such as running the VPN in a virtual machine or utilizing a cellular device's Wi-Fi for internet connection, the researchers from Leviathan Security indicate that no full-proof solution exists for other operating systems, highlighting a significant security gap in current VPN technologies.

Novel attack against virtually all VPN apps neuters their entire purpose Ars Technica