• AI KATANA
  • Posts
  • Meta’s AI Chatbot Achieves 400M Users, Closing Gap with ChatGPT

Meta’s AI Chatbot Achieves 400M Users, Closing Gap with ChatGPT

Also: Researchers Built an ‘AI Scientist’ — What Can It Do?

Author

AI KATANA
September 01, 2024

Morning!

In today’s AI-focused update, Meta's AI chatbot, powered by Llama 3.1, has achieved a remarkable milestone of 400 million monthly active users, drawing closer to ChatGPT’s dominance. Meanwhile, researchers have developed an "AI Scientist" capable of automating various aspects of the research process, though its current applications remain limited. Meta’s Llama 3.1 also continues to make waves in the open-source AI community, driving innovation across industries. Additionally, Microsoft has introduced its Phi-3.5 series of open-source models, further intensifying the competition in AI development. Finally, the U.S. AI Safety Institute has partnered with Anthropic and OpenAI to enhance AI safety protocols, a critical step in responsible AI advancement.

Sliced just for you:

  • 🤖 Meta’s AI Chatbot Achieves 400M Users, Closing Gap with ChatGPT

  • 🧠 Researchers Built an ‘AI Scientist’ — What Can It Do?

  • 🌐 Meta Llama 3.1: A Catalyst for Open Source AI Innovation

  • 💻 Microsoft Launches Open-Source Phi-3.5 Models for Advanced AI Development

  • 🇺🇸 U.S. AI Safety Institute Signs Agreements with Anthropic and OpenAI

  • 🕸️ AI Has Created a Battle Over Web Crawling

Meta's AI chatbot, powered by Llama 3.1, has achieved a significant milestone by reaching 400 million monthly active users across its integrated platforms, including Instagram, WhatsApp, and Facebook. This rapid growth reflects Meta's strategic push to expand its AI capabilities beyond the US, with the company aiming to make its chatbot the most used AI tool by the end of the year. However, despite these impressive figures, Meta's AI still trails ChatGPT, which boasts 200 million weekly active users. Meta's success can be attributed to the integration of its chatbot into popular services used by billions globally, though the company faces challenges with user retention and adoption in certain markets. Additionally, the open-source nature of Llama has driven substantial growth, with over 350 million downloads since its launch, although there are ongoing debates about the transparency of its training data.

A team of researchers from Sakana AI, along with collaborators from Canada and the UK, has developed an "AI Scientist," a large language model designed to automate the entire research process. This AI is capable of reading existing literature, formulating hypotheses, conducting experiments through algorithmic simulations, and even writing and reviewing its own research papers. Although the AI Scientist has demonstrated the ability to perform these tasks end-to-end, its current applications are limited to the field of machine learning, and it lacks the capacity to conduct laboratory work, which is considered a critical component of scientific research. The model's output so far has been incremental rather than groundbreaking, and it has drawn mixed reactions from the scientific community. Some researchers criticize its limited scope and tendency to favor popular research over novel insights, while others acknowledge its potential to automate repetitive aspects of research, allowing scientists to focus on more creative tasks. The developers view this as just the beginning, likening the current state of AI Scientist to the early stages of AI development, and anticipate significant advancements in the future.

Meta's Llama 3.1, an advanced AI model with 405 billion parameters, represents a significant leap in AI capabilities, excelling across diverse tasks such as complex mathematics, coding, and multilingual translations. With its open-source approach, Meta aims to democratize AI, similar to the impact of Linux in the software world. Since its release, Llama 3.1 has seen rapid adoption, with a 10-fold increase in usage from January to July 2024 and widespread implementation across various industries, including tech giants in China. Meta's partnerships with major cloud providers have further broadened Llama's reach. However, the open-source nature of Llama 3.1 has sparked debates about potential risks, such as misuse for malicious purposes. Despite these concerns, Meta remains committed to open-source AI, believing the benefits of widespread access outweigh the risks, positioning Llama 3.1 as a pivotal force in the future of AI innovation.

💻 Microsoft Launches Open-Source Phi-3.5 Models for Advanced AI Development

Microsoft has launched three new open-source AI models under the Phi-3.5 series, each designed to address different advanced AI tasks. The Phi-3.5-mini-instruct, with 3.82 billion parameters, is optimized for efficient reasoning and performs well in memory-constrained environments, even outperforming larger models like Meta's Llama-3.1-8B-instruct in some benchmarks. The Phi-3.5-MoE-instruct model, featuring 41.9 billion parameters and a mixture-of-experts architecture, excels in complex reasoning tasks, surpassing even Google’s Gemini 1.5 Flash in specific scenarios. The Phi-3.5-vision-instruct, with 4.15 billion parameters, integrates text and image processing capabilities, making it effective for multimodal tasks such as video summarization and optical character recognition. These models, trained on extensive datasets using powerful GPUs, demonstrate high performance across various benchmarks, positioning them as strong contenders in the AI landscape. Released under a permissive MIT license, these models offer developers the flexibility to innovate across diverse applications, from enterprise solutions to academic research. The Phi-3.5 series highlights Microsoft’s commitment to advancing open-source AI, offering scalable and efficient tools for a broad range of use cases.

The U.S. AI Safety Institute, part of the National Institute of Standards and Technology (NIST), has signed groundbreaking agreements with AI companies Anthropic and OpenAI to collaborate on AI safety research, testing, and evaluation. These Memoranda of Understanding grant the Institute access to new AI models from these companies before and after their public release, enabling joint research to assess capabilities, identify safety risks, and develop methods to mitigate those risks. This collaboration marks a significant step toward ensuring safe and trustworthy AI innovation, aligning with the Biden-Harris administration’s Executive Order on AI. The U.S. AI Safety Institute, in partnership with its U.K. counterpart, will provide feedback on potential safety improvements, furthering the global effort to responsibly advance AI technology.

The increasing use of robots.txt files by websites to block AI companies' web crawlers is creating significant challenges for the development of generative AI models, which rely heavily on vast amounts of public data for training. Traditionally used to manage web search engines' data collection, robots.txt has gained new relevance as more websites—particularly those concerned with monetization, like news sites and artists—restrict access to their content to prevent unauthorized use by AI systems. This growing trend is leading to a reduction in the availability of high-quality data, as recent studies have shown a notable rise in the percentage of data that has been restricted, particularly from major sources like news and academic sites. As AI companies face these new hurdles, they may turn to synthetic data or exclusive data agreements, though these approaches bring their own set of challenges, including potential legal disputes and concerns over data quality. This shift is sparking debates over the future of AI development, data accessibility, and the ethical implications of web crawling in the age of AI.

🧑🏽‍💻 AI Jobs

🛠️ AI tools updates

Apple has released the iOS 18.1 developer beta, introducing support for "Apple Intelligence," a suite of AI-powered features designed to enhance user experiences across iPhones, iPads, and Macs. This marks the first developer beta to integrate Apple Intelligence, following its announcement at the 2024 Worldwide Developers Conference. Key features include a redesigned Siri interface with keyboard interaction, AI-assisted writing tools that help with grammar, sentence restructuring, and summarization, as well as enhanced search capabilities within the Photos app. These AI features are currently available only to users in the U.S. running the latest iOS 18.1 developer beta on specific devices, such as the iPhone 15 Pro and Apple silicon-powered Macs and iPads. Users can access these tools by joining a waitlist through the Apple Intelligence & Siri menu in their device settings. Future updates are expected to unlock even more AI functionalities as Apple continues to develop its intelligence platform.

💵 Venture Capital updates

Swiss venture capital firm Redalpine has successfully raised a new $200 million fund, Redalpine Capital VII (RAC VII), to focus on early-stage investments in European startups. This fund was oversubscribed, highlighting the firm’s strong track record of delivering consistent top-quartile returns, with an average net return of around 25% per annum over the last decade. Redalpine, which manages over $1 billion in assets, is known for its focus on the intersection of software and science, particularly in deep tech. The firm’s strategy includes backing companies emerging from Europe’s university spinouts, leveraging its deep connections within the scientific community. RAC VII will continue this approach, with investments already made in companies such as Proxima Fusion, a German nuclear startup, and the U.K.’s ExpressionEdits, which uses AI to advance gene editing therapies. The new fund also supports Redalpine’s expansion, including plans to open a new office in London to tap into the region’s growing startup ecosystem.

🫡 Meme of the day

⭐️ Generative AI image of the day