OpenAI & Google Battle for AI Supremacy

Google's AI Avalanche: A Groundbreaking Week of Innovation

Everything Google Announced at Google I/O 2024

During the Google I/O Developer Conference 2024, Google unveiled a significant AI focus with Gemini enhancements, improved search faculties in Google Photos, and advanced AI integrations in Google Workspace. The update to Gemini, now with multimodality, lets it process various inputs and provide comprehensive outputs. The cloud-based Gemini 1.5 Pro offers developers greater computational power. Google Photos' new 'Ask Photos' employs AI for refined image searches. Google Workspace tools, like Gmail and Docs, now feature integrated Gemini AI for efficiency boosts, aiding in communication, documentation, and routine automation through a feature called Gems.

Additionally, Google introduced two new Gemini AI models: the speedy Gemini 1.5 Flash and Project Astra, a cutting-edge visual chatbot. Creative tools from Google Labs have also been upgraded, including VideoFX for generating videos from text prompts, ImageFX for improved image generation, and MusicFX's DJ Mode for creating music samples. Google Search underwent an evolution with new AI-organized search, multi-step reasoning, and advanced visual search in Lens. Security features were not left behind, as Google introduced an Android scam detection feature and expanded its SynthID watermarking tool to identify AI-generated content.

Sponsor

Vultr is empowering the next generation of generative AI startups with access to the latest NVIDIA GPUs.

Try it yourself when you visit getvultr.com/forwardfutureai and use promo code "BERMAN300" for $300 off your first 30 days.

• OpenAI’s Chief Scientist, a Former Foe of CEO Altman, Resigns with Key Ally - Ilya Sutskever, OpenAI's Chief Scientist and co-founder, announced his departure from the Microsoft-backed startup for a new, undisclosed project. His resignation follows the controversial firing and rehiring of CEO Sam Altman, which Sutskever was involved in. Jan Leike, another executive closely associated with Sutskever, also resigned. These departures are part of a broader executive turnover at OpenAI, including recent exits of other senior leaders and the firing of a team member for sharing information improperly. Despite this, OpenAI continues to recruit high-profile talent from leading tech companies.

• Meta Explores AI-Assisted Earphones With Cameras - Meta Platforms is exploring AI-powered earphones with cameras, aiming to identify objects and translate languages, though CEO Mark Zuckerberg has not approved any designs. Dubbed "Camerabuds," the project faces challenges like camera placement, potential overheating, and privacy concerns. Despite other tech companies pursuing similar AI wearables, the project’s timeline and final design remain unclear. Meta's history of struggling with hardware products raises questions about its potential success in the competitive market for smart earphones.

• Apple Launches New Accessibility Features Along with Eye Tracking Integration for iOS, iPadOS - Apple unveils an array of new accessibility features aimed at users with disabilities, including eye-tracking for iOS and iPadOS, allowing control using only eye movements, thanks to AI and the front-facing camera. The company also introduces Music Haptics, transforming audio cues into tactile feedback for the deaf and hard of hearing, compatible with Apple Music and available for developers via an API. Vocal Shortcuts and a function to assist users with speech conditions to interact with devices more efficiently are also highlighted, along with Vehicle Motion Cues to help with motion sickness by visualizing vehicle movements. These features are anticipated to launch with the upcoming iOS and iPadOS 18 updates later in the year.

• Hugging Face is sharing $10 million worth of compute to help beat the big AI companies - Hugging Face is investing $10 million in free GPUs for developers to foster new AI technology development and combat the centralization of AI progress in big tech companies. Their CEO, Clem Delangue, highlights that the substantial funding round the company received enables this initiative. Through the ZeroGPU program, Hugging Face provides shared GPU access via its Spaces platform to support applications and developers who need computational resources but face barriers with major cloud providers. The move aims to bridge the gap between proprietary and open-source AI, promoting a decentralized and collaborative future for AI and machine learning advancements.

  

• Google’s broken link to the web - At Google's I/O event, CEO Sundar Pichai introduced the concept of generative AI offering new opportunities across sectors. However, the event featured limited stage representation from developers and startups, centering more on celebrity engagements. Google showcased a variety of products, including AI tools designed to streamline everyday tasks and enhance Google services, such as organizing receipts, improving Google Photos queries, and creating interactive tutoring sessions with NotebookLM. A major highlight was the announcement of the Search Generative Experience, an AI-powered search tool set to launch in the U.S. and reach 1 billion users by the end of 2024. This tool represents a shift from web exploration to providing integrated answers, potentially decreasing outbound web traffic and impacting publishers—anticipated to result in significant revenue loss for content creators. Google's approach suggests a digital environment moving towards self-sufficiency, where the process of direct web browsing becomes less relevant as users are given AI-synthesized information. The potential implications of this evolution raise concerns among those reliant on traditional web traffic for their livelihood.

• Microsoft asks some China-based staff to consider relocating amid Sino-U.S. tensions - Microsoft is suggesting a voluntary transfer for some of its employees in China, specifically those involved in machine learning and cloud computing, amidst heightened tensions between the U.S. and China over advanced technology. The proposal, affecting an estimated 700–800 staff, comes as U.S. efforts to restrict China's access to sophisticated AI chips escalate, possibly impacting American businesses operating there. Despite this move, Microsoft asserts its commitment to continuing its operations in China. The employees have the option to relocate to countries including the U.S., Ireland, Australia, and New Zealand. This follows recent U.S. tariff increases on Chinese imports and potential new regulations on AI model exports.

• NASA Names First Chief Artificial Intelligence Officer - NASA has appointed David Salvagnini as the Chief Artificial Intelligence Officer to lead the agency's AI initiatives, emphasizing the responsible use of AI in space exploration and Earth applications. With a history of AI application, NASA aims to pioneer advancements, guided by ethical considerations, in alignment with President Biden's Executive Order on AI development. Salvagnini, with a background in intelligence and a 21-year Air Force career, will spearhead strategic AI planning at NASA, fostering collaborations to maintain cutting-edge AI capabilities. Previously, NASA's Chief Scientist Kate Calvin held the interim AI leadership role. Salvagnini’s role will support missions like analyzing Earth science imagery and managing data communication from extraterrestrial exploration like the Perseverance Mars rover.

• Insilico Medicine publishes CDK8/19 novel inhibitor powered by generative chemistry platform to treat multiple cancers - Insilico Medicine has developed a new drug that inhibits CDK8 and CDK19, which are enzymes involved in cancer cell growth and spread. This drug was created using AI technology, specifically a platform called Chemistry42, which generated over 3,600 potential molecules. The final compound shows strong effectiveness and good properties for being taken orally, with positive results in animal tests for treating various cancers. Insilico Medicine is a leader in using AI for drug discovery and has made significant advancements in treating diseases like cancer and fibrosis.

•  Instagram’s co-founder is Anthropic’s new chief product officer - Anthropic, an AI startup founded by former OpenAI employees, has appointed Instagram co-founder Mike Krieger as its chief product officer. Krieger's experience with intuitive products complements Anthropic's shift from focusing on core AI technology development to productization amidst a highly competitive AI industry. With recent ventures like the Claude app and multilingual support, Anthropic is poised to enhance user interaction, particularly in business contexts. The company is well-funded, boasting nearly $8 billion raised and backing from major investors like Amazon and Google, and is rumored to be collaborating with Apple. This strategic hiring signals Anthropics’ readiness to escalate their presence in the rapidly evolving AI market.

  

• Microsoft’s AI Push Imperils Climate Goal as Carbon Emissions Jump 30% - Microsoft's goal to be carbon negative by 2030 is jeopardized by its 30% increase in carbon emissions since 2020, driven by its aggressive expansion in artificial intelligence (AI). The company's AI advancements necessitate more energy-intensive data centers, which contribute significantly to its carbon footprint due to the use of cement, steel, and microchips. Despite this, President Brad Smith argues that the benefits of AI will outweigh its environmental impact and emphasizes the need for rapid progress in making AI more environmentally friendly. Microsoft's sustainability efforts include increasing energy efficiency, investing in green technologies, and reducing reliance on renewable energy credits (RECs) by focusing on power purchase agreements (PPAs).

  

• Senators Propose $32 Billion in Annual A.I. Spending but Defer Regulation - A bipartisan group of U.S. senators, led by Senator Chuck Schumer, has proposed a legislative framework for artificial intelligence (AI), seeking $32 billion in annual funding by 2026 to boost AI research and development. While the 20-page plan, “Driving U.S. Innovation in Artificial Intelligence,” acknowledges the need for a federal data privacy law and the regulation of deepfakes in elections, it provides limited detail on AI regulatory measures. The document calls for sector-specific agencies to develop regulations to prevent AI-related issues like job losses, health and financial discrimination, and copyright infringement. Schumer and his colleagues, following a yearlong inquiry into the AI field, have opted for a cautious approach, noting the fast-paced evolution of AI. This legislative action lags behind the European Union's comprehensive AI legislation, creating a disparity in AI governance between the U.S. and Europe.

Awesome Research Papers

What matters when building vision-language models? - This abstract discusses the critical need for justifying design choices in vision-language models (VLMs), highlighting a lack of rationale in many cases. The researchers address this by conducting comprehensive experiments on pre-trained models, architectures, data, and training methodologies. They consolidate their findings to develop Idefics2, an efficient VLM with 8 billion parameters. Idefics2 delivers state-of-the-art performance within its parameter range on multiple benchmarks, rivaling models four times larger. The team makes Idefics2 and its training datasets available in various versions, promoting transparency and potential advancement in the field.

MS MARCO Web Search: a Large-scale Information-rich Web Dataset with Millions of Real Click Labels - MS MARCO Web Search introduces a significant large-scale web dataset with millions of real clicked query-document labels, tailored to represent actual web search scenarios. It's designed to enhance research across multiple areas including neural indexer models, embedding models, and advanced information access systems incorporating large language models. The dataset is a pioneering resource, offering a retrieval benchmark through three web retrieval challenge tasks, and it demands innovative approaches in both machine learning and information retrieval. Providing a substantial contribution to AI and system research, the MS MARCO Web Search dataset is publicly accessible for research purposes.

Protecting users with differentially private synthetic training data - Google has developed methods for generating differentially private synthetic text, enabling the creation of data that maintains the essential characteristics of the original while providing robust privacy. Described in their work "Harnessing large-language models to generate private synthetic text," these methods ensure data privacy, making it suitable for diverse machine learning tasks including feature engineering, model selection, debugging, and monitoring. Google's synthetic data approach, useful for organizations lacking extensive ML expertise, facilitates privacy-preserving data sharing and has applications such as safe content classification.

Introducing PaliGemma, Gemma 2, and an Upgraded Responsible AI Toolkit - The Gemma family welcomes PaliGemma, an advanced open vision-language model (VLM) necessitating fine-tuning for various tasks. To support open research and collaboration, PaliGemma is accessible across multiple platforms, with Google Cloud credits available for academic research. The forthcoming Gemma 2 promises a more efficient architecture, delivering high performance with lower computational requirements. Additionally, the Responsible Generative AI Toolkit is being enhanced to facilitate safer and more responsible AI development.