June 05 2024 - June 12 2024
At its WWDC 2024 keynote, Apple unveiled Apple Intelligence, a suite of generative AI features integrated across iPhone, iPad, and Mac. Powered by a combination of on-device and cloud-based models, Apple Intelligence offers personalized intelligence while maintaining user privacy. Key features include systemwide writing tools for rewriting, proofreading, and summarizing text [1], a new generative image creation tool called Image Playground [3], and AI-assisted search and video editing in the Photos app [6].
Apple Intelligence also brings significant enhancements to Siri, making it more natural, contextual, and capable of understanding personal information across apps [12]. Users can now type to Siri in addition to voice commands [11]. Notably, Apple has partnered with OpenAI to integrate ChatGPT into Apple Intelligence, allowing users to access its language model through Siri and other apps [8]. However, Apple plans to support other AI models in the future, positioning ChatGPT as a potential interim solution [3].
While many Apple Intelligence features run on-device, Apple has introduced Private Cloud Compute for more complex tasks, ensuring user data is never stored on servers [12]. Independent experts can inspect the code running on Apple's servers, and devices will only communicate with servers whose software has been publicly logged [6]. Apple emphasized its commitment to privacy throughout the keynote, stating that Apple Intelligence "sets a new standard for privacy in AI" [12].
Although Apple Intelligence will be available as a free update, it will initially be limited to select devices with Apple's latest chipsets, including the iPhone 15 Pro series, iPads with M1 or newer chips, and Macs with M-series processors [7]. This hardware requirement highlights the computational demands of Apple's on-device AI approach [5].
Elon Musk's AI startup xAI announced plans to build the "world's largest supercomputer" in Memphis, Tennessee. The Greater Memphis Chamber revealed that the project, dubbed the "Gigafactory of Compute," is pending approval from authorities and is expected to create high-quality jobs. Memphis Mayor Paul Young stated, "We had an ideal site, ripe for investment. And we had the power of our people who created new and innovative processes to keep up with the pace required to land this transformational project." [5]
Reports emerged that Musk instructed Nvidia to redirect a shipment of AI chips originally intended for Tesla to xAI and X instead. Internal Nvidia emails obtained by CNBC showed that the move delayed over $500 million worth of chips to Tesla by months. Musk clarified on X that Tesla had no place to put the chips where they could be turned on, and they would have sat idle in a warehouse. [9]
xAI raised $6 billion in a Series B funding round, valuing the company at $24 billion. The funding included $750 million from Musk himself and $250 million worth of computing power from X. [1] Additionally, the Xai Foundation partnered with Outlier Ventures to launch a bootcamp aimed at helping Web2 developers transition to Web3 gaming and blockchain development for the Xai network. [3]
Kuaishou Technology, a major Chinese tech company, has unveiled an AI video generator called Kling that can produce remarkably realistic videos up to two minutes long from text prompts. Kling has quickly gained attention for its impressive capabilities, which rival OpenAI's much-anticipated Sora model. [1] Kling can generate 1080p videos at 30 frames per second, adopting a "3D spatio-temporal joint attention mechanism" that enables accurate modeling of complex movements and adherence to the laws of motion. [1] Users have shared examples showcasing Kling's abilities, such as generating videos of a man eating noodles, a cat driving a car, and various imaginative scenarios. [5]
Kling's immediate availability through a waitlist in China has allowed users to explore its practical applications, demonstrating its potential in creating immersive visual experiences. [6] The model supports various aspect ratios and utilizes advanced 3D face and body reconstruction to enhance facial expressions and limb movements. [2] Kling has been described as a formidable rival to Sora, surpassing it in terms of video length, resolution, and accuracy in certain aspects. [10] However, global accessibility remains a potential hurdle for Kling, as China has been reluctant to grant worldwide access to its cutting-edge technologies. [11]
The emergence of Kling has positioned China as a leader in the field of text-to-video generation, joining other Chinese models like Vidu AI and Shengshu Technology's Vidu. [5] These developments represent a significant milestone for China's technological advancement and underscore the global competition driving rapid innovation in AI. [6] While OpenAI has announced plans to launch Sora later this year, the race for AI supremacy is heating up, with Kling offering a glimpse into China's technological prowess. [6] "Kling is the new text-to-video model launched by Kuaishou Technology, a Chinese company known for its short-video platform similar to TikTok." [10]
Alibaba has recently made significant strides in the field of large language models (LLMs), unveiling its latest open-source model series, Qwen2. According to Alibaba, Qwen2-72B, the high-end variant, consistently outperformed Meta's Llama 3-70B across various benchmarks, including mathematics, coding, natural and social sciences, engineering, and the humanities. [1] The Qwen2 series ranges from 0.5 billion to 72 billion parameters and has been pre-trained on data covering 27 languages in addition to Chinese and English. [2]
Notably, Qwen2-72B-Instruct achieved impressive results in the "Needle in a Haystack" test, demonstrating its ability to extract relevant information from large contexts without errors almost perfectly. [2] Moreover, the Qwen2 models boast an expanded context window of up to 128K tokens, matching the capabilities of OpenAI's GPT-4o. [6] Alibaba has also adopted the Apache 2.0 license for most Qwen2 models, aligning with standard open-source software practices. [2]
The release of Qwen2 follows Alibaba's previous announcement of its closed-source Tongyi Qianwen 2.5 model, which was noted for its superior Chinese language capabilities compared to GPT-4. [3] Alibaba aims to leverage AI, including Qwen2, to transform businesses and boost efficiency across various sectors. [1] The company has positioned itself as a pivotal player in the global AI landscape, fostering innovation and broader usage of AI technologies through open-sourcing its models. [5]