• Bits & Neurons
  • Posts
  • Google’s Gemini AI won’t be available in Europe — for now

Google’s Gemini AI won’t be available in Europe — for now

📰 Quick News 📰

😔 Early impressions of Google's Gemini, a new generative AI model, are disappointing. Users have reported that the model fails to provide accurate answers, struggles with translation, and is unreliable for summarizing news. It also lacks coding proficiency and is susceptible to jailbreaks. Google claimed that Gemini Pro, the lite version, exceeded the performance of other gen AI models, but it seems to have some limitations. Overall, Gemini's early impressions suggest that there is room for improvement in its functionality and performance.

+ Gemini will not be available in Europe & UK due to regulatory hurdles, for now.

🌍 EU negotiations on the first major AI law hit a snag but continue Friday; Companies in Europe using powerful AI models will face new EU regulations, documenting their AI's abilities and risks. The law targets AI with "systemic risk," which includes AI trained on supercomputers with significant power. Open-source AI models are exempt, for now.

⚙️ AMD has released new chips, the Instinct MI300X accelerator and the Instinct M1300A accelerated processing unit (APU), for running large language models (LLMs) and training AI. These chips have improved memory capacity and are more energy-efficient than previous versions. AMD also announced the Ryzen 8040 processor for mobile devices, which offers enhanced AI processing performance.

+ Meta, OpenAI, and Microsoft are planning to use AMD's new AI chip, the Instinct MI300X, as an alternative to Nvidia's graphics processors for AI programs. This move could lower costs and put competitive pressure on Nvidia's AI chip sales.

🦿 Apple quietly unveiled MLX, a tailor-made machine learning framework for developers using Apple Silicon, aiming to facilitate the creation, training, and deployment of AI models. The framework is engineered to work efficiently with Apple's own processors and is equipped with features that accelerate computation and ease development in multiple languages. MLX is not meant for average consumers but for developers and researchers.

📚 Nerd section 📚

🖥️ Long context prompting for Claude 2.1 - Claude 2.1 has a 200,000 token context window and performs well on real-world retrieval tasks. However, it can be reluctant to answer questions based on individual sentences that are out of place. By using a minor prompting edit, Claude 2.1's performance on these tasks significantly improves.

⚒️ How Tiny LLMs are Redefining the AI Landscape - lightweight models with fewer parameters are challenging the narrative that bigger models are always better in AI. They prioritize efficiency and specialized functionality over sheer scale.

📖 Calling an AI prompt 'You are an expert' does not make it knowledgeable. It's a misconception that doesn't enhance the AI's knowledge. The prefix doesn't impact the chatbot's response quality.

🌐 Web section 🌐 

🕸️ What's this new section about?

On days when there's a lot of news, we aim to keep our newsletter short and sweet.

So here (but not in the email version) we're adding AI-related stories that may be slightly less attention-grabbing, but still could move the needle in specific sectors and fields.

😢 Sports Illustrated fired two high-ranking officials following a controversy involving articles authored by AI-generated fake writers. This issue arose amid falling stock prices for the company, with claims from the owners that the firings were part of a broader restructuring rather than being solely related to the AI scandal, which had already caused significant damage to the brand's reputation and market performance.

👨 Google cofounder Sergey Brin has reportedly played a key role in the creation of Gemini, the company's latest AI model. This confirms reports that Brin has been more actively involved at Google as it tries to catch up to its AI rivals. Brin is listed as a "core contributor" to Gemini and has been seen frequently working at Google's headquarters.

🎨Meta launched an AI tool that transforms text into images, but it produces low-quality results compared to others like Dall-E, even though it's free with unlimited use.

🕷️ Almost every Windows and Linux device vulnerable to new LogoFAIL firmware attack - Security flaws in the firmware of Windows and Linux devices allow for the execution of malicious code during the boot-up sequence, bypassing traditional defense mechanisms and gaining high levels of control over the devices.

(Not AI news, but important for all of us anyway)

💰 Five-month-old Indian AI startup Sarvam raises $41 million from investors including Lightspeed, Peak XV, and Khosla Ventures. Sarvam is building large language models with support for Indian languages and aims to create a platform for businesses to build with these models. The startup plans to make its first model public in the coming weeks.

🍔 McDonald's is partnering with Google to introduce AI in its stores for better operations and fresh food. Updates will occur in 2024, affecting thousands of locations, potentially impacting labor dynamics.

Thank you for reading today’s edition!

We love to hear back from you!

Feel free to reply to our emails with questions, suggestions, or topics you'd like to see covered, or drop us a message on Twitter or Facebook.

Until tomorrow,
- Ts (Bits and Neurons)