Good morning! Groq is challenging NVIDIA's AI chip supremacy, a YouTuber is taking OpenAI to court, and Amazon's bold 'DoctorAI' move in healthcare.

It’s the Tuesday edition, folks, let’s dive in (no splash).

Stories You Should Know:

• NVIDIA in the Hot Seat: NVIDIA's dominance in the AI chip market has caught the DOJ's attention with not one, but two antitrust probes.

• Google Cloud's GPU Giveaway: Y Combinator startups hit the jackpot with a dedicated cluster of NVIDIA GPUs, courtesy of Google Cloud.

• Intel's Epic Fall: Once the king of Silicon Valley, Intel's stock has nosedived, leading to massive layoffs.

• Britain’s AI Setback: The UK government has canceled £1.3 billion in AI computing projects, dampening its AI ambitions.

• CISA's AI Power Move: The Cybersecurity and Infrastructure Security Agency (CISA) has named Lisa Einstein as its first Chief Artificial Intelligence Officer.

Sponsor

CodiumAI is a quality-first generative AI coding platform, offering developers tools for writing and refactoring, as well as testing and reviewing. Generate confidence, not just code. Try for free.

• AI Chip Startup Groq Valued at $2.8 Billion After Latest Funding Round - Groq, a Silicon Valley-based semiconductor startup, has raised $640 million in a Series D funding round led by Cisco Investments, Samsung Catalyst Fund, and BlackRock Private Equity Partners, elevating its valuation to $2.8 billion. Specializing in AI inference chips, Groq has adapted Meta's large language model, LLaMA, to operate on its chips, competing with NVIDIA's dominance in the AI chip industry. The funds will support the expansion of Groq's tokens-as-a-service offering and the deployment of over 108,000 Language Processing Units by early 2025. Additionally, Groq has appointed Stuart Pann as COO and Yann LeCun as a technical adviser.

• YouTuber Files Class Action Suit Over OpenAI’s Scrape of Creators’ Transcripts - YouTuber David Millette has filed a class action lawsuit against OpenAI, alleging the company used millions of YouTube video transcripts to train its AI models without notifying or compensating creators. The complaint, filed in the U.S. District Court for the Northern District of California, accuses OpenAI of violating copyright laws and YouTube's terms of service. The lawsuit seeks over $5 million in damages for affected YouTube users and creators. This legal action comes amidst increasing difficulties for AI companies in accessing high-quality training data, with many websites blocking OpenAI's data collection efforts.

• Elon Musk Revives Lawsuit Against OpenAI and Sam Altman - Elon Musk has reignited a longstanding dispute by re-filing a lawsuit against AI company OpenAI, accusing its founders, Sam Altman and Greg Brockman, of prioritizing commercial success over the public interest, which contradicts the company's original mission. Musk’s allegations highlight a shift away from OpenAI's commitment to developing AI for the common good following a lucrative deal with Microsoft. Musk's action comes after an earlier withdrawal of the lawsuit just before a judicial decision on its dismissal was anticipated. OpenAI has not yet issued a response to the new legal challenge, while its founders previously indicated an intent to seek dismissal of Musk's original claims.

• Amazon has discussed a 'DoctorAI' tool to automate routine healthcare tasks - Amazon's One Medical team aims to revolutionize healthcare efficiency with their proposed "DoctorAI" large language model. This innovation, discussed in a December 2023 internal document, seeks to automate routine tasks for clinicians, enhance product and service recommendations, and manage customer service interactions. Amazon is positioned uniquely for this venture, leveraging its access to first-party health data from One Medical and Amazon Pharmacy, and expert advice from its large clinical panel. The company's current AI capabilities are underscored by in-house models like Titan and Olympus, fueling its commitment to the generative AI market. Amazon projects a minimum of $1 billion in revenue from AI initiatives in the current year, as stated by CEO Andy Jassy.

  

• CISA Names First Chief Artificial Intelligence Officer - The Cybersecurity and Infrastructure Security Agency (CISA) has appointed Lisa Einstein as its first Chief Artificial Intelligence Officer, highlighting its commitment to leveraging AI for cybersecurity and infrastructure protection. Einstein, who has been leading CISA’s AI initiatives since 2023, brings a wealth of experience and leadership to the role. CISA Director Jen Easterly praised Einstein's contributions and leadership, emphasizing the importance of AI in enhancing CISA’s cyber defense capabilities. Einstein expressed her dedication to ensuring AI's safe and secure deployment to protect critical systems and infrastructure in the United States.

Sponsor

AI Hub by Qualcomm - Run, download, and deploy your optimized models on Snapdragon® and Qualcomm® devices.  Learn more about AI Hub by Qualcomm at https://aihub.qualcomm.com/

• NVIDIA faces two DOJ antitrust probes over market dominance - The US Department of Justice (DOJ) is conducting two investigations into NVIDIA’s business practices related to its AI division. The first probe is examining the implications of NVIDIA's acquisition of Run:ai, an Israeli startup focused on GPU management software, within the context of antitrust laws. The second inquiry is delving into accusations that NVIDIA has misused its control over the AI chip market, which it dominates by an estimated 70-95%, to limit customers' use of competitors' products and to enforce pricing strategies that disadvantage customers who choose rival chips. These legal scrutinies come amid broader regulatory attention on tech giants for potential anti-competitive conduct, with particular concern about the AI sector. NVIDIA asserts its commitment to legal compliance, innovation, and supporting choice in the market. This follows reports of possible antitrust charges against NVIDIA in France and the company's recent market valuation achievements, bolstered by the high demand for AI technologies.

• Google Cloud now has a dedicated cluster of NVIDIA GPUs for Y Combinator startups - Google Cloud is providing Y Combinator startups with a dedicated, subsidized cluster of NVIDIA GPUs and Google TPUs to facilitate AI model development, along with $350,000 in cloud credits over two years. This initiative aims to build long-term relationships with early-stage AI startups, encouraging them to grow within the Google Cloud ecosystem. The partnership offers YC startups priority access to compute resources, essential support credits, and direct interaction with Google's AI experts. This move aligns with a broader trend among tech giants and venture capital firms to attract and support promising AI startups by providing substantial computing power.

  

• Britain cancels $1.7 billion of computing projects in setback for global AI ambitions - The UK government has canceled £1.3 billion in computing infrastructure projects, undermining efforts to become an AI leader. The scrapped initiatives include a £500 million investment in the AI Research Resource and an £800 million project for an exascale computer at the University of Edinburgh. The Labour government's reallocation of funds reflects a shift in priorities, seeking economic stability over the previously funded AI commitments. Fiscal prudence is emphasized as the government faces significant previous unfunded pledges. Moves towards AI infrastructure enhancement continue with the launching of an AI Opportunities Action Plan, although the introduction of a U.K. AI Bill was notably absent from a recent speech by King Charles III, indicating a careful approach towards regulating the AI industry.

• Intel was once a Silicon Valley leader. How did it fall so far? - Intel's stock has plummeted to its lowest level since 2013, as the company announced a substantial 15% workforce reduction following a major dip in revenue and a $7 billion loss in its chip foundry business for 2023. This marks Intel's largest downturn in five decades. The company is implementing cost-saving measures including layoffs and suspended stock dividends, aiming to save $10 billion by 2025. Once a semiconductor industry pioneer, Intel struggles to keep pace with competitors such as NVIDIA, despite efforts to innovate with its Gaudi technology. Intel's future hinges on the profitability of its foundry business and its positioning in an industry facing both excitement and apprehension about the rapid growth of AI technology.

  

• Big Tech Fails to Convince Wall Street That AI Is Paying Off - Amazon, Microsoft, and Alphabet have disappointed investors despite significant investments in AI technology, as their recent earnings reports show inadequate financial returns from these efforts. While there was some growth in their cloud-computing divisions, it was not sufficient to satisfy Wall Street's expectations. Amazon's and Microsoft's stocks have both declined following their earnings announcements, and Alphabet's stock fell by 7.4%. Meta and Apple, however, have shown more promising results with their AI investments, highlighting a varied response among tech giants regarding AI profitability.

• NVIDIA’s New AI Chip Delayed Due to Design Flaws: Information - NVIDIA's latest AI chips in the Blackwell series will face delays of three months or more due to design flaws, pushing big shipments to the first quarter of 2025. This setback, reported by The Information, affects major customers like Meta, Google, and Microsoft, who were recently informed about the issue. NVIDIA has stated that production is still on track to ramp up later this year, though specific comments on the delay were not provided. Major tech companies involved declined to comment on the situation.

• AI music startup Suno claims training model on copyrighted music is 'fair use' - Suno, an AI music startup, has admitted to training its AI model using copyrighted songs, asserting that this practice is legal under the fair-use doctrine. The admission follows a lawsuit filed by the Recording Industry Association of America (RIAA) against Suno and Udio, alleging unauthorized use of copyrighted material. Suno's CEO compared the process to learning music by listening to it, while the RIAA argues that such practices constitute large-scale infringement. The case, still in its early stages, is expected to set a significant precedent regarding the application of fair use in AI model training.

  

• Character.AI CEO Noam Shazeer returns to Google as the tech giant invests in the AI company - Noam Shazeer, co-founder and CEO of Character.AI, is rejoining Google after initially leaving to establish the chatbot startup backed by a16z. Along with co-founder Daniel De Freitas and other employees, Shazeer will join Google DeepMind, while Dominic Perella steps in as interim CEO at Character.AI. Google has signed a non-exclusive agreement with Character.AI to use its technology, providing the startup with continued funding. This move may attract regulatory scrutiny from bodies like the FTC and EU, similar to recent investigations into tech acquisitions.

• There’s a Tool to Catch Students Cheating With ChatGPT. OpenAI Hasn’t Released It - OpenAI has developed a tool that can detect AI-generated text with 99.9% accuracy, but internal debates over potential impacts have delayed its release. Concerns include possible disproportionate effects on non-native English speakers and the risk of students evading detection by manipulating the text. While some employees argue for the tool's benefits, including assisting educators in curbing academic dishonesty, others worry about negative user reactions and the potential impact on ChatGPT's adoption. The company's leadership has prioritized transparency and responsible AI use, but the ultimate decision on the tool's release remains unresolved.

  

• Leaked Documents Show NVIDIA Scraping ‘A Human Lifetime’ of Videos Per Day to Train AI - Internal documentation and communications acquired by 404 Media reveal NVIDIA's development of an unreleased video foundational model. The nature of the model and its applications are not specified in the brief content provided. Further details about the project's scope, its potential impact on NVIDIA's product lineup, or how it may influence the video technology landscape are currently unavailable in the summarized content.

• Mercedes approved for L4 driving tests in Beijing, ahead of Tesla’s robotaxi - The Mercedes-Benz Group has become the first international auto manufacturer to receive approval for level-4 (L4) autonomous vehicle testing in Beijing, allowing it to conduct road tests which include complex maneuvers without human intervention. These trials, highlighting advancements in sensor technology and safety, mark a significant step since Beijing began drafting supportive self-driving regulations. Previously, only local Chinese companies had received such permissions. The city has designated over 1,160 kilometers of roads for testing, involving 384 vehicles from 18 companies. Meanwhile, Tesla is still awaiting a similar permit, following CEO Elon Musk's lobby for a robotaxi service in China. This development reflects the broader race amongst Chinese cities to commercialize autonomous driving technology.

  

• Max’s new homepage personalization tech makes viewers stick around, says exec - HBO Max, recently rebranded as Max, has implemented a "whole page optimization" system in the U.S. that personalizes the entire homepage for users, contributing to increased viewer engagement. Warner Bros. Discovery reports extended viewing duration, higher return visits, and diversified content consumption as results of the new feature. The algorithm tailors the homepage based on individual viewing habits, offering targeted collections such as "heart-pounding thrills" or "supernatural scares." Unlike the former method that combined human curation and limited personalization, this full-page approach refines content visibility and variety. Future plans include a Netflix-like rating system for user feedback and AI integration for emotionally contextual recommendations, underscoring the significance of personalization in retaining subscribers in the competitive streaming landscape.

• ‘You are a helpful mail assistant,’ and other Apple Intelligence instructions - Apple's developer betas introduced AI features expected to roll out to devices in the coming months, showcasing AI's expanding role in the user experience. Users within the macOS 15.1 beta community discovered underlying prompt files for these AI features, revealing the directions guiding the AI's behavior, such as remaining concise, avoiding fabrication, and steering clear of sensitive topics. While these files underpin features like Smart Replies and the Photos app's "Memories" video creation, they are uneditable and underscore the controlled parameters guiding Apple's AI tools to ensure appropriate user interactions. This discovery also hinted at the internal workings labeled "ajax," presumably Apple's LLM (large language model), offering a peek into AI's operational framework within Apple's ecosystem.

• OpenAI Tempers Expectations with Less Bombastic, GPT-5-less DevDay this Fall - OpenAI announced a shift in its DevDay conference format, opting for smaller, regional developer sessions rather than a major event, and confirmed it will not unveil its next flagship model at the event. Instead, the focus will be on updates to APIs, developer services, and educational content for developers. This change comes as OpenAI faces challenges with data access for training its models, leading to incremental improvements rather than major advancements. Despite this, the company remains under pressure to deliver high-performance AI models amidst ongoing controversies and financial strain.

  

Awesome Research Papers

• Towards Achieving Human Parity on End-to-end Simultaneous Speech Translation via LLM Agent - This paper introduces the Cross Language Agent -- Simultaneous Interpretation (CLASI), a simultaneous speech translation system with a data-driven read-write strategy for high-quality translation and low latency, mirroring professional interpreters. CLASI uses multi-modal retrieval to handle in-domain terminologies and leverages large language models (LLMs) to produce error-tolerant translations based on audio, historical context, and additional information. It achieves substantially higher valid information proportion (VIP) scores, conveying information effectively in real-world scenarios with informal and disfluent speech, outperforming commercial and open-source counterparts by a significant margin.

• Berkeley Humanoid: A Research Platform for Learning-based Control - Berkeley Humanoid introduces a mid-scale, cost-effective humanoid robot crafted for research in learning-based control. It stands out for its straightforward design tailored for algorithms that minimize simulation complexity, emulate human-like movement, and exhibit high resilience to falls. Its design substantially narrows the sim-to-real gap, allowing for agile, robust mobility on various outdoor terrains using a basic reinforcement learning controller with minimal domain randomization. The robot has demonstrated capabilities such as prolonged traversing, navigating steep unpaved paths, and single and double leg hopping, highlighting its dynamic walking proficiency. Emphasizing scalability, the Berkeley Humanoid supports omnidirectional movement and can recover from significant disturbances, advancing the practical deployment of learning-based humanoid technologies.

THUDM/CogVideoX-2b - Presenting an open-source video generation model called CogVideoX, which derives from the 清影 project. CogVideoX-2B, the current model version, requires 36GB of GPU memory for inference and supports English prompts with a length of 226 tokens to generate 6-second videos at 8 frames per second and a resolution of 720 * 480. It utilizes sinusoidal positional embeddings, doesn’t support quantized inference or multi-card inference, and advises using an SAT model for reduced memory use. The platform offers quick deployment guidelines using the Huggingface diffusers library, and a demonstration showcases the model's capability to create a video of a panda playing guitar in a bamboo forest.

Announcing Flux by Black Forest Labs: The Next Leap in Text-to-Image Models - Black Forest Labs has unveiled Flux, the most substantial open-source text-to-image model to date featuring 12B parameters. Highlighted for its impressive creativity and performance, Flux offers three variants on the fal platform: FLUX.1 [dev] with non-commercial licensing, FLUX.1 [schnell] for up to 10 times faster operation, and the API-exclusive FLUX.1 [pro]. These variants aim to produce highly realistic and anatomically precise images with a remarkable adherence to prompts. Enhanced by fal's inference engine, Flux models can deliver images at twice the speed of standard methods, combining expedited processing with high quality. Users can experience Flux through demos or explore more via fal Playgrounds and API documentation.

Introducing torchchat: Accelerating Local LLM Inference on Laptop, Desktop and Mobile - Team PyTorch introduces torchchat, a new library to efficiently run Llama 3 and other large language models on various devices such as laptops, desktops, and mobile phones. Utilizing PyTorch 2, torchchat enhances performance across different environments, and supports functionalities like model export, quantization, and evaluation. It offers a Python-based REST API, a C++ desktop binary, and ExecuTorch for mobile on-device inference. Performance metrics reveal impressive speeds for the Llama 3 model on devices with varying hardware configurations and data types, with particularly high tokens per second rates on Apple's M1 Max and using CUDA on Xeon CPUs with A100 GPUs.

Introducing Hero - Hero is a new voice-guided assistant designed to simplify daily life by integrating various functions into one platform. Powered by Google Calendar, Apple Weather, Perplexity, Instacart, and Reminders, Hero allows users to manage their schedules, tasks, and communications with voice commands, eliminating the need to switch between multiple apps. Created for busy individuals, couples, and parents, Hero aims to streamline processes and reduce the hassle of coordinating activities. Currently available on iPhone in the US and Canada, with plans for an Android version and global availability soon.

Prompt Design at Character.AI - Prompt Poet, introduced by Character.AI, is a novel tool designed to enhance the process of prompt engineering, shifting it towards a more efficient and user-friendly method of prompt design. It appeals to both technical and non-technical users by utilizing Python f-strings encapsulated within a flexible YAML and Jinja2 templating system, allowing sophisticated prompt creation without the need for coding. The Prompt Poet Library incorporates features like tokenization and cache-aware truncation for optimized responses.

Google’s NeuralGCM is available on GitHub - NeuralGCM is a Python library for building hybrid ML/physics atmospheric models for weather and climate simulation.