May #TechBytes - 1st part

Here is a roundup of some of the tech news you may have missed recently selected by us at Etiqa!

Ours is a journey inside the latest advanced technologies, news about the most exciting startups and scale ups from around the world, and updates on specialized, high-risk industries.

Microsoft ventures into new AI territory with MAI-1

At the end of this month, 微软 might unveil a powerful new generative AI model dubbed MAI-1, potentially rivaling OpenAI with its capacity to handle around 500 billion parameters, as reported by The Information.

When it comes to investing in artificial intelligence, Microsoft stands out.

According to a report from The Information, despite having already invested $10 billion in OpenAI , the Redmond giant is now working on a new generative AI model, named MAI-1. This endeavor aims not only to challenge Alphabet (Google's parent company) but, ironically, also its own partner, OpenAI.

The project is reportedly led by Mustafa Suleyman, co-founder of Google DeepMind and former CEO of the startup Inflection AI, which gave birth to the "empathetic chatbot" Pi.

In addition to Suleyman, Microsoft has reportedly hired a significant portion of Inflection AI's employees and paid a hefty $650 million to acquire materials from the latter for use in part of MAI-1's training process (as per The Information, the other training data would come from OpenAI's GPT-4 model).

Once MAI-1's training is complete, it is expected to handle around 500 billion parameters, a significant feat considering that, according to unofficial estimates, the GPT-4 model can handle 220 billion parameters trained on 8 separate weight sets for a total of 1760 billion parameters.

The exact reason why Microsoft is venturing into such powerful AI territory remains unclear. Nevertheless, we might find out soon, as per the aforementioned report, Microsoft could unveil MAI-1 at the upcoming Microsoft Build conference, scheduled from May 21st to 23rd. When contacted by Reuters, Microsoft declined to comment on the news.

TikTok introduces AI to detect AI-generated content

With numerous elections on the horizon, TikTok is taking steps to combat the spread of deepfakes and other non-real content by utilizing Adobe's Content Credentials technology. This move marks TikTok as the first video upload platform to implement such measures.

Artificial intelligences present both benefits and challenges in our modern world, particularly in the realm of information dissemination and social media, where discerning truth from falsehood becomes increasingly complex.

As various elections, from European to American presidential races, loom in 2024, platforms are intensifying efforts to counter synthetic media—content created or modified with AI—either by labeling it or restricting its spread. TikTok stands out among these platforms, not only enforcing rules for artificial videos for over a year but now also being the pioneer in using AI to identify them.

Adobe's Content Credentials, the chosen tool, is significant in that it has become somewhat of an industry standard. Similar to Instagram's potential future use for images, TikTok will utilize Adobe's AI to identify and label AI-generated content (AIGC). What sets this initiative apart is its application not only to videos created within TikTok but also those produced elsewhere. For instance, if a video is edited using tools like CapCut and then uploaded to TikTok, even if it appears as authentic content, TikTok's platform will recognize it as AI-generated.

Adam Presser, responsible for Operations & Trust and Safety at TikTok, confirmed this expansion, stating, "This allows us to extend automatic labeling to AIGC created on some other platforms, a capability we plan to progressively apply to all users globally in the coming weeks."

TikTok has formed a significant partnership with the Coalition for Content Provenance and Authenticity (C2PA), uniting leading tech companies to establish standards for AI use. Presser added, "In the coming months, we will also begin attaching Content Credentials to TikTok content, where they will remain even after downloaded." This attachment of metadata ensures that the AI Generated label remains even if the content is downloaded and reused.

Additionally, TikTok is collaborating with MediaWise to introduce new features aimed at media literacy, specifically addressing AI. These efforts reflect TikTok's commitment to transparency and authenticity in the face of increasingly sophisticated synthetic media.

While TikTok's measures represent a significant step forward, they are not a panacea for all challenges posed by AI. Nonetheless, they mark a crucial development, especially in an election year, and will likely gain traction as more platforms adopt Content Credentials, making AI-generated content increasingly recognizable and labelable.

iOS 18: Apple and OpenAI collaborate on new AI features

According to recent rumors circulating the web, 苹果 is reportedly teaming up with OpenAI to develop AI-driven functionalities. Here's what we know so far:

As widely speculated, Apple is expected to officially unveil new AI-related features at the Worldwide Developers Conference 2024 (WWDC 2024). The event is anticipated to showcase the latest functionalities arriving with iOS 18 and the new iPhone 16, marking Apple's entry into the AI sector and competing with already established players in the field.

Additionally, Bloomberg's Mark Gurman suggests that Apple has been engaged in intensified discussions with OpenAI, indicating a potential collaboration between the two companies, possibly involving the integration of a version of ChatGPT into Apple's devices.

Details regarding the agreements between Apple and OpenAI remain somewhat elusive. However, rumors suggest that Apple has been in talks with various AI companies for several months, aiming to establish partnerships to bring AI functionalities to its upcoming products.

While specifics of the collaboration are still unclear, there are indications of two main avenues of exploration: an on-device AI developed by Apple, prioritizing user privacy and potentially exclusive to flagship devices equipped with the A18 Bionic processor featuring an updated Neural Engine; and a cloud-based AI solution, possibly a customized version of ChatGPT, catering to older devices compatible with iOS 18 that lack the hardware requirements for Apple's proprietary AI project.

Further speculation hints at ongoing discussions between Apple and 谷歌 , potentially leading to the introduction of a version of Google Gemini tailored for iPhones.

Moreover, reports from Mashable suggest that Apple researchers are working on an AI update for Siri, enhancing the capabilities of the voice assistant and integrating it with a new LLM (Language Learning Model) to improve user interaction and on-screen comprehension.

While these developments remain speculative, all eyes will be on Apple's announcements at WWDC 2024 in June to confirm the extent of their progress in the realm of artificial intelligence.

Google introduces advanced A.I. for understanding human molecules

Artificial intelligence continues to revolutionize various fields, from generating videos to accelerating scientific research aimed at understanding the complexities of the human body and combating diseases.

Recently, Google DeepMind, in collaboration with Isomorphic Labs, unveiled an enhanced version of AlphaFold, an AI technology designed to aid scientists in comprehending the behavior of microscopic mechanisms within human cells.

The original AlphaFold, launched in 2020, successfully tackled a long-standing scientific challenge known as "the protein folding problem." Proteins, essential molecules in all living organisms, undergo intricate folding processes that determine their functions. AlphaFold proved instrumental in rapidly predicting the three-dimensional shapes of proteins, a task that previously consumed significant time and resources.

Since its release, AlphaFold has been pivotal in expediting drug discovery efforts. Researchers worldwide have utilized this technology to delve into various medical inquiries, including studies on COVID-19, malaria, and Parkinson's disease.

The latest iteration, AlphaFold3, extends its capabilities beyond protein folding to predict the behavior of other biological components like DNA and RNA. This expansion enables a deeper understanding of the dynamic interactions between different molecules within biological systems.

Dr. John Jumper, a researcher at Google DeepMind, emphasized the profound insights AlphaFold3 offers into cellular mechanisms and disease processes. This advancement represents a significant leap forward in understanding biological systems and holds immense potential for accelerating drug development and vaccine creation.

AlphaFold3 is accessible to scientists through a dedicated website, providing a powerful tool for advancing research and potentially saving months of experimental work. Its unparalleled accuracy, as demonstrated in a recent paper published in the scientific journal Nature, marks a significant milestone in the field of computational biology.

Demis Hassabis, CEO of Google DeepMind and founder of Isomorphic Labs, underscores the importance of understanding molecular interactions in biology. AlphaFold3 represents a crucial step towards unraveling the complexities of biological systems and facilitating groundbreaking research previously deemed unattainable.