• AI Breakfast
  • Posts
  • Meta's AI generates images from brain data in milliseconds

Meta's AI generates images from brain data in milliseconds

Good morning. It’s Friday, October 20th.

In today’s email:

  • Partnerships and Collaborations for AI Advancement

  • AI Model Development and Advancements

  • Research and Development in AI Technology

  • Corporate Business and Financial Impacts of AI

  • AI Applications for Improved User Experience

  • 5 New AI Tools

  • Latest AI Research Papers

You read. We listen. Let us know what you think of this edition by replying to this email, or DM us on Twitter.

Today’s edition is brought to you by:

A Banksy got everyday investors 32% returns?

Mm-hmm, sure. So, what’s the catch?

We know it may sound too good to be true. But thousands of investors are already smiling all the way to the bank, thanks to the fine-art investing platform Masterworks.

These results aren’t cherry-picking. This is the whole bushel. Masterworks has built a track record of 15 exits, including net returns of +10.4%, +27.3%, and +35.0%, even while financial markets plummeted.

But art? Really? Okay, skeptics, here are the numbers. Contemporary art prices:

  • outpaced the S&P 500 by 131% over the last 26 years

  • have the lowest correlation to equities of any asset class

Got your attention yet? AI Breakfast readers can skip the waitlist with this exclusive link.

Today’s trending AI news stories

Partnerships and Collaborations for AI Advancement

Nvidia and iPhone maker Foxconn to build 'AI Factories' leveraging Nvidia chips to power various applications, including autonomous vehicles and large language models. Amidst US plans to restrict advanced chip exports to China, the partnership aims to revolutionize manufacturing, emphasizing the emergence of intelligence-driven data centers. The move marks Foxconn’s transition to a platform solution company, while Nvidia’s stock value exceeds $1 trillion.

Amazon and MIT are partnering to study how robots impact jobs The collaboration with Ipos research firm examines the impact of robotics and AI on jobs. The study aims to understand the implications for human workers and public perception as automation increases in industrial settings. With over 750,000 robots already in use, Amazon emphasizes the importance of technology supporting safety and improving employee tasks. MIT’s research highlights the importance of effective teamwork between humans and robots.

Universal Music strikes ‘first-of-its-kind’ strategic AI partnership with BandLab Technologies The collaboration aims to empower emerging artists within BandLab’s global community and promote the ethical uses of AI while safeguarding artist and songwriter rights. BandLab Technologies, the parent company of BandLab, has a powerful AI-driven music creation platform with over 60 million registered users, making it the world’s largest social music creation platform.

Infosys and Google Cloud expand alliance to help enterprises transform into AI-first organizations Infosys will establish Generative AI Labs and train 20,000 practitioners on Google Cloud’s generative AI technologies, fostering AI integration into business processes. The collaboration aims to enhance existing platforms such as Infosys Live Enterprise Application Management and Infosyst Customer Intelligence with Infosys Topaz and Google Cloud’s generative AI capabilities. This partnership will empower enterprises to achieve AI-enabled transformation and drive long-term digital success.

AI Model Development and Advancements

[Top Story]

Meta's new AI system can generate images from brain data in milliseconds Meta AI has developed a system using magnetoencephalography (MEG) to decode visual representations in the brain, potentially paving the way for non-invasive brain-computer interfaces. The real-time image reconstruction sheds light on how the brain processes images, contributing to Meta AI’s goal of developing AI systems that mimic human learning and reasoning. The research offers insights into the foundations of human intelligence.

OpenAI reportedly canceled "Arrakis", its more efficient GPT-4 level AI model The cancellation was reportedly due to issues with its quality - Arrakis utilized the sparse principle, activating only parts of the neural network. Despite starting development in the fall and commencing training in the spring, performance fell short. OpenAI might integrate Arrakis’ work into other models like the Gobi multimodal model. In response to the setback, OpenAI refocused on creating a GPT-4 version with faster response generation.

Fuyu-8B: A Multimodal Architecture for AI Agents Adept introduces Fuyu-8B, a small version of its open-sourced multimodal model tailored for digital agents. Fuyu-8B’s simplified architecture allows for easy understanding, scaling, and deployment. With its cavity for fine-grained image processing and fast response times, the model showcases impressive performance on image-based benchmarks. Its capabilities span understanding images, charts, documents, and diagrams, while upcoming releases promise added functionalities for OCR, UI element localization, and question-answering about UIs.

AI company Anthropic sued for copyright infringement by Universal Music Group Anthropic, recently backed by Amazon, is facing a copyright infringement lawsuit from Universal Music Group, and ABKCO. Allegedly, Anthropic’s AI model “Claude” scraped copyrighted song lyrics without permission for its content creation.

AI tidies up Wikipedia’s references — and boosts reliability The SIDE system, developed by Samaya AI, employs a neural network to identify and replace unreliable citations. Trained on reputable Wikipedia articles, it verifies reference quality and suggests alternatives from the internet. Approximately 21% of users preferred AI-suggested citations, emphasizing the potential time-saving benefits for editors. However, user preferences highlight the need for careful implementation and community acceptance.

‘Mind-blowing’ IBM chip speeds up AI IBM’s NorthPole processor represents a groundbreaking advance in AI computing, integrating memory and processing to enhance speed and reduce power consumption. By eliminating the Von Neumann bottleneck, it outperforms existing AI machines in image recognition while consuming just one-fifth of the energy. While not yet suitable for large language models, NorthPole’s design principles offer promise for speed-critical applications like autonomous vehicles. Innovative approaches using new materials and manufacturing processes may further enhance computing efficiency.

Research and Development in AI

Microsoft wants to make AI safer, and it just unveiled a service to help The company has launched Azure AI Content Safety, an AI-powered platform designed to create a safer online environment by detecting inappropriate content. Originally part of the Azure OpenAI Service, it’s now a standalone system, available for open-source and other company models. Azure AI Content Safety allows businesses to tailor policies, aligning content with their values. The platform has previously been used in Microsoft’s own AI-powered products, and now it’s available for external organizations, allowing them to customize policies.

Nvidia bets on generative AI for robotics and industrial applications The company unveiled significant advancements in its Jetson platform for edge AI and robotics, introducing generative AI for industrial applications and robotics, enabling learning with limited or no examples. The platform also integrates natural language interfaces for streamlined AI application development. The company launched the Jetson Generative AI Lab, offering open-source models, and updated robotics and video AI frameworks for simplified development.

Corporate Business and Financial Impacts of AI

OpenAI in talks to sell employee shares at $86 billion valuation OpenAI is negotiating the sale of employee shares at an $86 billion valuation, positioning itself as one of the world’s most valuable private companies. With Microsoft owning 49% of OpenAI, the company expects to generate $1.3 billion in annual revenue, driven primarily by ChatGPT Plus subscriptions. However, despite significant revenue, OpenAI may not achieve profitability soon due to high operational costs for large language models.

The world’s biggest AI models aren’t very transparent, Stanford study says The Foundation Model Transparency Index by Stanford examines the disclosure practices of major AI models, revealing a lack of transparency among leading developers such as OpenAI and Meta. The report outlines 100 indicators of transparency, with Meta’s Llama 2 scoring highest, followed by BloomZ and OpenAI’s GPT-4. Notably, none of the models provide information on societal impacts, raising concerns about privacy and bias. Policymakers may use the index as a benchmark for future regulations.

AI Applications for Improved User Experience

After 50,000 hours, this AI can play Pokémon Red The reinforcement learning algorithm, developed by Peter Whidden, has fascinated millions with its gameplay. While unable to interpret in-game dialogue, the AI exhibits quirky behaviors, such as pausing to admire scenery and forming negative associations, presenting a unique exploration of AI behavior through the lens of the iconic Pokémon game.

5 new AI-powered tools from around the web

Jumble Journal is an empowering app fostering clarity through distraction-free journaling. Equipped with progress tracking and secure zero-knowledge architecture, users delve into introspection, organization, and insight generation, nurturing their potential with peace of mind.

Sales Sparrow, an open-source AI-powered Salesforce companion, streamlines note-taking and task management post-sales calls. Seamlessly integrating with Salesforce, it offers AI-recommended next steps, ensuring no crucial details slip through the cracks. Easy accessibility and task synching enhance sales efficiency.

Impaction.ai offers the ultimate solution for analyzing conversational AI products. It presents real-time data on an intuitive dashboard, enabling easy natural language searches and analysis with Columbus. It aims to enhance LLM-powered products. Receive insights to optimize user interactions efficiently.

Webstudio AI is an innovative co-pilot for web designers, that streamlines website development through voice-activate prompts. Functioning as an aid, not a replacement, it empowers designers to generate content and execute complex CSS tasks effortlessly. With an open-source approach, it promises a collaborative user-centric web design experience.

Middleware is an advanced cloud observability platform that empowers businesses with comprehensive insights into complex applications. Its AI-driven capabilities facilitate issue detection and resolution, ensuring real-time data access and robust security.

arXiv is a free online library where scientists share their research papers before they are published. Here are the top AI papers for today.

This paper introduces “MusicAGent,” a sophisticated AI system designed for seamless music processing, encompassing tasks from music generation to understanding, powered by LLMs. With a comprehensive array of music tools, the system organizes user requests, selects suitable tools, and generates comprehensive responses, simplifying complex music-related tasks. Supporting text, sheet music, and audio formats, MusicAgent ensures accessibility, unity, and modularity in music processing. It streamlines the integration of various tools from multiple sources, including Hugging Face, GitHub, and web APIs, facilitating an enhanced and efficient music experience. By leveraging the capabilities of LLMs, MusicAGent revolutionizes the process of handling intricate music tasks, catering to a diverse audience with varying levels of expertise.

The paper introduces a novel framework that enhances LLMs with adaptive retrieval and self-reflection for improved factual accuracy without compromising versatility. The proposed Self-Reflective Retrieval Augmented Generation (Self-RAG) framework dynamically retrieves and generates text while self-evaluating its outputs using special tokens. The approach enables the model to tailor its behavior to diverse task requirements, leading to significant improvements in various downstream tasks, including open-domain QA, reasoning fact verification, and long-form generations. The framework outperforms existing models on these tasks, ensuring improved factuality, citation accuracy, and overall generation quality.

The paper introduces LAMP, a few-shot text-to-video generation framework aimed at balancing generation freedom and training costs. Leveraging a first-frame-conditioned pipeline, LAMP decouples content and motion, utilizing a T2I model for first-frame generation and subsequent motion learning layers capture temporal features, while the shared-noise sampling strategy stabilizes the generation process. LAMP outperforms existing methods in terms of alignment, consistency, and diversity, demonstrating its effectiveness in learning motion patterns from limited data. User studies confirm LAMP’s superiority in visual quality and motion matching.

The paper introduces a framework that leverages LLMs for instruction-driven 3D modeling. This addresses the complexity of procedural generation in 3D content creation, enabling more efficient and precise modeling. The framework involves three core agents: the task dispatch agent, the conceptualization agent, and the modeling agent. These agents collaborate to interpret instructions, enrich scene descriptions, and generate Python code for controlling Blender’s 3D modeling. The experiments demonstrate the framework’s proficiency in generating 3D scenes that align with user instructions, highlighting its potential in simplifying and enhancing the 3D modeling workflow.

The paper introduces an approach, AutoMix, that strategically leverages diverse black-box Large Language Models (LLMs) for cost-effective and accurate problem-solving. It employs a self-verification mechanism and a meta-verifier to determine when to route queries to larger models. The method introduces a Partially Observable Markov Decision Process (POMDP) to categorize queries as Simple, Complex, or Unsolvable, optimizing computational resource allocation. AutoMix achieves an incremental benefit per cost (IBC) improvement of up to 89%, surpassing FrugalGPT in various datasets. The POMDP-based meta-verifier consistently demonstrates superior performance, ensuring reliable decision-making.

Thank you for reading today’s edition.

Your feedback is valuable.

Respond to this email and tell us how you think we could add more value to this newsletter.