Welcome back to AI Hungry, where the pulse of artificial intelligence innovation and regulation beats strong. In our latest roundup, we delve into the controversy surrounding Google's new AI system, Gemini, which has come under fire for not living up to its hyped capabilities. Additionally, we explore the European Union's groundbreaking AI Act, set to establish a global benchmark for AI governance with its comprehensive and stringent regulations.
These stories highlight the dynamic tension between AI's rapid advancement and the imperative for responsible oversight. As the AI landscape continues to evolve, stay informed with our in-depth coverage of these pivotal developments.
Google's New AI System Gemini Faces Criticism Over Exaggerated Capabilities
Google has introduced Gemini, an artificial intelligence system designed to rival OpenAI's GPT-4. The system was presented as a highly capable digital assistant, able to interact with visual data and solve problems in real-time. However, the tech community has raised concerns that Gemini's actual performance does not match Google's bold claims, particularly after early testers found it struggling with basic tasks.
Tech experts and early users have criticized Google for potentially overstating Gemini's abilities, with some pointing out inconsistencies between the company's benchmarks and the AI's real-world performance. The revelation that Google's promotional video was pre-recorded, rather than a live demo, has added to the skepticism surrounding the AI's capabilities, highlighting the gap between marketing and actual technology.
European Union Finalizes Landmark AI Act to Set Global Standards
The European Union has reached a consensus on the AI Act, a comprehensive set of regulations for artificial intelligence, aiming to serve as a model for global legislation. The law, which is not the first of its kind but is the most extensive to date, includes prohibitions on AI systems that identify individuals by sensitive traits, restrictions on mass facial data collection, and stringent rules for powerful AI models. Companies that fail to adhere to these regulations could face fines up to 7% of their global turnover. The prohibitions will be effective in six months, transparency requirements in a year, and the complete set of rules within approximately two years.
The AI Act also introduces measures to protect copyright holders from generative AI and mandates transparency about AI systems' energy consumption. The legislation reflects a shift in focus due to the rapid evolution of AI technology and its applications, as evidenced by the release of OpenAI's ChatGPT. The final negotiations included heated debates on the use of biometric identification by law enforcement and the regulation of foundational AI models versus their applications. Despite some concerns about potential loopholes for law enforcement, the EU has taken a significant step towards preemptively managing the risks associated with AI, contrasting with its reactive approach to social media regulation in the past.
🤖 French Startup Mistral AI Unveils Groundbreaking Mixtral 8x7B Language Model. French startup Mistral AI has launched Mixtral 8x7B, a new language model rumored to follow an architecture similar to OpenAI's GPT-4. The model, released via torrent, has not yet been benchmarked but is expected to build on the success of its predecessor. (Link)
🚀 AMD Challenges NVIDIA with New High-Performance AI Accelerators. AMD has released the MI300X and MI300A accelerators, marking a significant advancement in the AI and HPC sectors. These accelerators, built on advanced architecture, are poised to challenge NVIDIA's market dominance with superior memory capacity and efficiency. (Link)
🤖 Elon Musk Launches Grok: A New AI Chatbot for Premium+ Subscribers. Elon Musk's company has launched Grok, an AI chatbot, for Premium+ subscribers in the US, with plans to expand globally. Despite being in beta with expected issues, Grok offers unique real-time data interactions, aiming to enhance the subscription model amid advertising challenges. (Link)
🚀 StripedHyena: A Groundbreaking AI Model Surpassing Traditional Transformers. Together AI unveils StripedHyena, a new AI model with 7 billion parameters, promising faster performance and efficiency over traditional transformers. It features a state space model layer, enabling it to process up to 128,000 tokens effectively. (Link)
🧠OpenChat: A Breakthrough in AI Language Models with Mixed Data Quality. Researchers have unveiled OpenChat, a new AI framework that enhances language models using mixed-quality data. Its unique C-RLFT method classifies data without explicit labeling, simplifying implementation and reducing costs. Early tests show OpenChat outperforming larger models and rivaling ChatGPT. (Link)
🧬 FDA Approves First Gene Editing Treatment for Sickle Cell Anemia. The FDA has approved a groundbreaking gene editing therapy for sickle cell anemia using CRISPR technology. While not a cure, it could replace bone marrow transplants and is particularly significant for the Black community, which is disproportionately affected by the disease. (Link)
