Perplexity AI, a company led by CEO Aravind Srinivas, is planning to develop a pure AI audio device for under $50. This device will allow users to ask questions and receive spoken answers. Srinivas describes Perplexity not as a search engine but as an answer machine. The development of this device is contingent upon receiving over 5000 likes on a social media post, which it achieved. However, details about the device’s release date or additional features remain unclear. Many major AI companies are currently announcing plans to enter the consumer hardware market. OpenAI is collaborating with former Apple designer Jony Ive on a new device, Meta offers a voice-controlled AI glasses, and Google is working on smart glasses as well.
Concerns are rising about the influence of AI-generated fakes on voters. According to AI researcher Gerard de Melo from the Hasso-Plattner-Institut, misinformation and manipulation via social media using AI could increase, especially during elections in Germany. The biggest worry is that people might receive completely false information and images generated by AI. The German Federal Office for the Protection of the Constitution warns of potential foreign interference in upcoming elections using AI to create deepfake videos and clone voices quickly. Although it’s uncertain if this poses a significant threat for the upcoming elections, de Melo expects more verification mechanisms to ensure human authenticity and to expose bots.
A research team has developed a benchmark platform called BALROG to test large language and visual language models in various gaming environments. The tests range from simple tasks to complex games like the “NetHack Learning Environment.” Models such as GPT-4o, Claude 3.5, and Llama 3.2 were evaluated. Results showed significant limitations in current AI language models, with top models like GPT-4o scoring only 32% on average. All models struggled with complex games requiring long-term planning. The findings highlight the need for improvements, especially in decision-making based on visual information and applying abstract knowledge to concrete situations.
Intellivision Technologies claimed its facial recognition technology was highly accurate and unbiased. However, the US Federal Trade Commission (FTC) is addressing misleading advertisements, as Intellivision could not substantiate these claims. The company allegedly trained its system on only about 100,000 real faces, with the rest being AI-generated variations. To avoid legal proceedings, Intellivision agreed to stop making misleading claims about accuracy, fake face detection, or bias comparisons.
Amazon introduced new foundational AI models called “Nova” for text, image, and video analysis. These models are divided into understanding and creative models. Nova Pro, the most powerful understanding model, processes text, image, or video inputs, generates text outputs, and can compete with other models. It handles up to 300,000 tokens and complex workflows involving APIs and external tools. A more advanced model, “Nova Premier,” is set for release in early 2025. For creative applications, Amazon offers Nova Canvas for image generation and Nova Reel for video production, initially available exclusively through Amazon Web Services in the US.
Tencent unveiled HunyuanVideo, a new open-source model for AI-driven video generation. With over 13 billion parameters, it’s the largest publicly available model of its kind, according to Tencent. The model surpasses existing systems like Runway Gen-3 and Luma 1.6, especially in motion quality. It can generate videos from text, convert images to videos, and create avatar animations. Audio generation for videos is also included. Tencent released it as open source on GitHub to bridge the gap between closed and open systems, with plans for continuous development.
Nextcloud Talk, a collaboration and video conferencing software, received a significant update named Paris. It now includes a desktop client for Windows, macOS, and Linux. New features include enhanced video conferencing and webinar capabilities, and a chat and meeting AI assistant. This AI can automatically create transcripts from recorded video conferences and summarize chat messages, helping users catch up quickly after absences. Nextcloud aims to offer an open-source alternative to Microsoft Teams with this update.
Generative AI is impacting the job market more significantly than previous automation processes. According to a report by the Organization for Economic Co-operation and Development (OECD), regions and sectors previously unaffected by technology are now experiencing automation of cognitive and creative tasks. The report highlights the influence on jobs in finance, insurance, communication, and technology. These jobs are primarily located in urban areas, shifting the impact of automation from rural to urban settings. While rural jobs like farming are less affected, urban areas like Berlin could see up to 70% of jobs influenced by AI, compared to 40% in regions like Thuringia. Despite these changes, researchers do not expect an overall decline in job numbers.