• AI Breakfast
  • Posts
  • AI Designs New Robots From Scratch in Seconds

AI Designs New Robots From Scratch in Seconds

Plus, watch as AI is integrated into a robot and interviewed on TV

Good morning. It’s Wednesday, October 4th.

Did you know: SkyNews Australia recently interviewed Amica, a GPT-powered robot bust with dynamic facial features.

In today’s email:

  • Enterprise AI Integrations

  • AI in Social Media and Advertising

  • Investments in AI Companies

  • AI in Research and Academic Applications

  • Technological Advancements in AI

  • Regulatory Developments

  • 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:

Fed up with the endless hunt for royalty-free music for your content?

Splash Pro is the most powerful AI music engine on the planet, letting you describe the song you want to hear, customize it, download it and deploy it in any project you choose!

AI Breakfast readers can use the code OGHMUXXD for 50% off our MAX tier for 2 months!*

  • Unlimited generative songs

  • 15 x AI singers & rappers

  • Production-quality mastering

  • Unlimited commercial license

*Offer applies to ‘MAX’ subscription tier & expires 30 Dec 2023

Today’s trending AI news stories

Enterprise AI Integrations

LinkedIn is introducing a suite of new AI-powered features across its platform, including AI assistance in recruiting, an AI-driven learning coach, and a tool for AI-enhanced marketing campaigns. The platform, with nearly 1 billion users and $15 billion in revenues last year, is leveraging technology from OpenAI, in which Microsoft owns a significant stake. The new features include AI-assisted recruiting with generative AI, a learning coach chatbot focused on soft skills, AI-enhanced marketing with a product called Accelerate, and AI support for B2B selling.

Microsoft has integrated OpenAI’s DALL-E 3 model into Bing Chat and Bing Image Creator, offering users the ability to generate images through conversation with a chatbot. DALL-E 3 is designed to better understand prompts and create more creative and photorealistic images. Microsoft has also incorporated safety features to avoid generating images of public figures or NSFW content. This integration makes Microsoft one of the popular image-generation tools. OpenAI plans to use DALL-E tech in other applications, such as the Paint Cocreator tool in the Windows Paint app.

Arc Browser is incorporating AI features called Arc Max, powered by GPT-3.5 and Anthropic. These features include Ask ChatGPT for queries from the Arc Command Line, Tidy Tab Titles for easy tab identification, Tidy Downloads for renaming downloads, Five-Second Previews for summarizing links, and Ask on Page for webpage-specific queries. These user-centric AI tools enhance the browsing experience without overwhelming users with excessive AI integration. The browser company will let users vote on which features to keep, prioritizing simplicity and user preference in AI adoption.

AI in Social Media and Advertising

MrBeast calls out TikTok for allowing a deep fake version of him hawking $2 iPhones TikTok removed the ad and the associated account for policy violations. The incident highlights the need for transparency in AI-generated content featuring celebrities, as unauthorized AI-generated content featuring public figures becomes more common on social media platforms. Tik Tok currently prompts uploaders to tag whether or not the video shorts contain AI generated images, and declares the right to remove any violations of this policy.

Researchers Tested AI Watermarks—and Broke All of Them A recent study reveals significant vulnerabilities in AI watermarking techniques designed to identify AI-generated content. Researchers found that both “low perturbation” and “high perturbation” watermarks can be evaded and even faked, posing challenges in detecting deepfakes and manipulated content. Despite efforts by major tech companies like Google, experts caution that watermarking may not be a foolproof solution. While some see it as part of a broader detection strategy, others argue that its real-world applications are limited. The study highlights the need for more effective methods to combat AI-generated content manipulation.

Investments in AI Companies

Anthropic, a startup competing with OpenAI, is reportedly in talks to raise over $2 billion in funding from Google. This comes shortly after Amazon committed to investing $1.25 billion in the company. Google, which previously acquired around 10% of Anthropic in 2022, is expected to participate in this funding round. Anthropic aims for a valuation between $20 billion to $30 billion, significantly increasing its worth compared to a $4 billion valuation just months ago. The company is known for its chatbot, Claude, which rivals OpenAI’s ChatGPT.

AI in Research and Academic Applications

AI beats human sleuth at finding problematic images in research papers AI tools are proving more effective than humans at identifying problematic images in research papers. An independent biologist, Sholto David, compared his manual examination of papers with an AI tool called ImageTwin. The AI outperformed David, identifying almost all the suspicious papers he had flagged and even finding additional problematic images he had missed. AI tools like ImageTwin are being adopted by universities, publishers, and scientific societies to tackle image manipulation in scientific papers, which can sometimes occur unintentionally. However, experts note that human expertise is still crucial in detecting fraudulent alterations.

Instant evolution: AI designs new robots from scratch in seconds The AI system developed by Northwestern University, which runs on a lightweight personal computer, generates wholly novel robot designs without relying on human creativity or large datasets. The researchers tested the AI by asking it to design a robot capable of walking on land. The AI quickly iterated on the design, eventually creating a robot that could walk its body length per second in just 26 seconds. The research could lead to the development of AI-designed tools with various applications.

Watch the video here:

Technological Advancements in AI

Tachyum is set to construct a 50 exaFLOP supercomputer powered by its 5nm Prodigy Universal Processor chip for a US client. This supercomputer is anticipated to deliver over 50 exaFLOP performance, which is 25 times faster than current systems, and will support AI models that are potentially 25,000 times larger. The Prodigy chip enhances memory, storage, and compute architectures, serving data center, AI, and HPC workloads across various sectors. Installation is expected to begin in 2024, providing significant AI training capabilities and computational power.

Regulatory Developments

Stricter EU controls on critical technologies possible from spring 2024 The European Commission plans to initiate collective risk assessments, in collaboration with member states, on four critical technology areas: advanced semiconductors, artificial intelligence, quantum technologies, and biotechnologies. These assessments, set to be completed by year-end, may lead to measures like export controls or development support by spring 2024. The move aims to address technology security risks and potential misuse. The EU also lists six more technology areas for discussion, emphasizing its goal of “de-risking” trade relationships rather than “de-coupling them, with China likely to be a key focus. However, divergent member state interests may pose challenges.

5 new AI-powered tools from around the web

TMate is an AI-driven meeting assistant that enhances productivity by transcribing and extracting valuable insights from meetings. It offers AI-generated summaries, highlights, and analytics, empowering users to make informed decisions efficiently.

Intently converts LinkedIn actions into sales opportunities, offering a streamlined approach to finding ready-to-buy leads. It enables efficient lead generation by identifying prospects and discussing relevant business problems on the platform, saving time and effort in the process.

Wois is an innovative platform for personal brand building, offering an asynchronous audiovisual experience. It enables users to speak freely, connect with professionals globally, and generate AI-powered social media content from their discussions.

Visily 2.0 revolutionizes wireframe and prototype creation, catering to users with no design skills. It offers AI-powered features such as UI generation from different sources, UI presets, and easy customization. Visily 2.0 also introduces text-to-design AI and enhanced collaboration tools, making it a comprehensive solution for teams and individuals in UX/UI design.

ChatDB is an AI-powered database assistant that streamlines data retrieval from databases, eliminating the need for complex SQL queries. It offers visualizations, table viewing, CSV manipulation, SQL formatting, and more. With technical support and customizable plans, ChatDB simplifies data management.

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

SmartPlay introduces a challenging benchmark and methodology for evaluating large language models (LLMs) as intelligent agents. It comprises six different games, including Rock-Paper-Scissors, Tower of Hanoi, and Minecraft, each testing specific capabilities of LLMs. SmartPlay aims to assess skills such as reasoning with object dependencies, spatial reasoning, and learning from history. This benchmark provides a standardized environment for evaluating LLM agent performance and identifying research gaps. It follows a unified interface with text-based observations and manuals and can be used to evaluate various LLMs.

The paper introduces LLM-grounded Video DIffusion (LVD), a training-free approach for text-to-video generation. LVD leverages Large Language Models (LLMs) to generate dynamic scene layouts aligned with text prompts. These layouts guide a video diffusion model, significantly improving video quality and alignment with desired attributes and motion patterns. The paper proposes a benchmark for evaluating generated videos and highlights the LLM’s ability to generalize and adapt to different scenarios. LVD offers a promising solution for enhancing text-to-video generation by harnessing LLM’s capabilities to generate realistic videos from intricate text prompts.

The paper critically examines the concept of self-correction in Large Language Models (LLMs) with a focus on reasoning tasks. While LLMs have shown promise in various applications, concerns about their accuracy and appropriateness persist. The paper explores self-correction, where LLMs refine their responses based on feedback, and highlights its limitations in intrinsic self-correction, where external feedback is absent. Empirical analysis indicates that LLMs struggle to improve their reasoning without external guidance, often degrading performance. The study also discusses the use of multi-agent debate and self-consistency methods, which improve performance but do not necessarily reflect genuine self-correction in LLMs.

The paper introduces a unified framework to assess the performance of various conditional image generation models, addressing inconsistencies in experimental conditions, including datasets, inference methods, and evaluation metrics. The paper presents ImagenHub, which includes standardized datasets, a unified inference pipeline, and human evaluation metrics for semantic consistency and perceptual quality. The study evaluates around 30 models across seven tasks and highlights several key takeaways, such as the generally unsatisfying performance of existing models and the need for continuous evaluation and progress tracking in this rapidly evolving field.

The authors propose a novel prompting approach called analogical prompting. Inspired by human analogical reasoning, this approach guides large language models (LLMs) to self-generate relevant exemplars or knowledge before solving a given problem. It eliminates the need for manual labeling of exemplars, offering adaptability and generality. Experimental results demonstrate that analogical prompting outperforms existing methods in various reasoning tasks, including math problem-solving and code generation. This approach improves LLM’s performance by providing tailored exemplars and knowledge, making it a promising direction for enhancing reasoning capabilities.

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.