Neuralink, Deepmind, and MechGPT

Author

AI Breakfast
November 27, 2023

Good morning. It’s Monday, November 27th.

In Partnership with CodeMode: Validate your business plan with AI today.

In today’s email:

  • Advancements in AI Technology

  • AI Development and Global Competition

  • AI Security and Regulations

  • 5 New AI Tools

  • Latest AI Research Papers

You read. We listen. Let us know what you think by replying to this email.

Interested in reaching 45,223 smart readers like you? To become an AI Breakfast sponsor, apply here. 

Today’s trending AI news stories

Advancements in AI Technology

> Neuralink secures an additional $43 million in funding for brain-implant tech, aiming to decode brain waves. The venture, detailed in an SEC filing, marks the Musk-led startup’s strides in developing implantable chips. Musk also eyes India for Tesla and Starlink expansions post-meeting with PM Modi.

> Voice AI reshapes real-time voice modulation, offering thousands of options for streamers, gamers, and content creators. The platform’s unique features include voice cloning, custom integration, and soundboards, providing an extensive user-generated content voice library, Voice Universe. With its roots in synthetic speech tech, Voice AI aims to redefine expression through audio, catering to diverse applications beyond gaming and content creation.

> Google DeepMind introduced Step-Back Prompting, elevating large language model accuracy by up to 36%. This technique involves posing a general question before the main task, allowing better comprehension and retrieval of background data. Tests on language models PaLM-2L and GPT-4 showed improvements, particularly in multi-step reasoning tasks across physics, chemistry, and temporal knowledge-based questions, surpassing GPT-4 in accuracy.

> MechGPT, crafted by MIT’s Markus J. Buehler, is a specialized language model designed for material failure research. Trained through question and answer pairs extracted from technical texts, it connects disparate research domains, helping in hypothesis generation. Buehler highlights MechGPTs broader applicability in virtual labs, teacher-student dialogues, and industries. Interaction tips underscore precise prompts and step-by-step explanations, while future prospects include combining MechGPT with few-shot learning and multi-agent strategies.

> University of Chicago researchers unveiled the 3D Paintbrush, an AI method generating local stylized textures on meshes using text input. Controlled by free-form text, this tool produces texture maps for meshes, enhancing graphics pipelines. Employing cascaded score distillation (CSD), their method refines details and region resolution. The approach combines neural fields and MLPs for precise texture and localization, providing high-resolution, coherent, and less noisy textures.

AI Development and Global Competition

> In a bid to evade Nvidia’s GPU embargo in China, numerous factories are reconfiguring GeForce RTX 4090 gaming cards into AI accelerators. Chinese facilities are dismantling these coveted GPUs and refitting them onto specialized PCBs with compact blower-style coolers, transforming them into high-demand AI tools. As the ban hampers accessibility, these factories offer an alternative route for Chinese consumers, amplifying demand and price hikes in the market.

> French AI lab Kyutai, supported by prominent billionaires including Xavier Niel, obtains a €300 million budget for open-source AI research. Recruiting former Meta FAIR and Google DeepMind researchers, Kyutai plans to release its debut model in a year. During the ai-PULSE conference, President Macron defended France's stance on the EU AI law, promoting open-source strength. Xavier Niel echoed support, highlighting Europe's AI innovation lag and regulatory concerns.

> ByteDance integrated AI into its collaboration tool, Feishu 7, introducing the Feishu Intelligent Buddy. This AI-powered assistant generates emails, analyzes documents, and aids in task automation, aligning with China’s tech giants’ push for industrial generative AI use. Users can choose from various large language models (LLMs), such as Tencent-backed MiniMax and state-backed Zhipu AI. Feishu already finds utility in companies like Anker and Yuangi Senlin.

> Wipro partners with NVIDIA AI Enterprise to propel healthcare insurance into the future with generative AI. This collaboration aims to enhance member experience, claims adjudication, and enrollment processes in ACA, Medicare, and Medicaid. Leveraging NVIDIA's AI Enterprise software, Wipro seeks to introduce AI-driven solutions fostering improved productivity, deeper member intelligence, and operational efficiency within healthcare organizations.

AI Security and Regulations

> The United States, Britain, and several nations have introduced a foundational international agreement that aims to safeguard AI systems against malicious use. While non-binding, the 18-country accord emphasizes monitoring AI for misuse and securing data against tampering, marking a milestone in recognizing the criticality of AI security during the design phase.

> Vladimir Putin declared plans for a national AI strategy to counter Western AI dominance, fearing Western-biased models that could potentially erase Russian culture. Putin outlined this strategy at the AI Journey conference in Moscow. He criticized existing models for favoring English-based data, proposing Russia expand its supercomputing, revise laws, and enhance international cooperation in AI.

> The Pentagon’s strategic use of AI now extends to roles like piloting tiny surveillance drones and aiding Ukraine against Russia. This AI also predicts maintenance needs for Air Force jets and monitors adversaries’ space satellites. Eyeing the Replicator initiative, the Defense Deputy Secretary urges rapid adoption of small, smart, cost-efficient autonomous vehicles by 2026.

In partnership with CODEMODE

Transform Your Business Idea into Success with AI-Powered Idea Validation

Fast-track your entrepreneurial journey with CodeMode’s efficient, AI-driven platform. Validate your business idea in minutes (not weeks) at a fraction of the cost.

Key Benefits:

Quick Validation: Get actionable insights swiftly.

Easy to Use: User-friendly interface, no jargon.

Comprehensive Insights: From industry trends to competitor analysis, CodeMode has you covered.

Download a personalized report tailored to your industry and audience today

Thank you for supporting our sponsors!

5 new AI-powered tools from around the web

MyMap MindMap Generator simplifies ideas visualization for study or work projects. It’s a free, user-friendly tool founded by Luo from EarlyBird, enabling effective learning and planning.

Albus for Community is an AI-powered Slack bot that transforms communities into knowledge hubs. Instant answers, learning from Slack convos, and multiple app integrations.

Insanely Fast Whisper utilizes OpenAIs Whisper Large V3 for swift audio transcription. CLI script, inference API, optimizations like batching and flash attention accelerate the process. Includes Roadmap and Community showcase.

Vsub is a video captioning tool with 98.5% accuracy in minutes. Features auto-generated captions, highlighted keywords, trendy templates, and animated emoji captioning. Streamlines content creation, boosts engagement for creators.

AiTerm is an AI Terminal Assistant streamlines natural language to executable commands. Offers on-demand documentation, personal assistance, and interactive CLI for developers, enhancing terminal efficiency.

arXiv is a free online library where researchers share pre-publication papers.

Memory Augmented Language Models through Mixture of Word Experts (MoWE) introduces a new approach merging Mixture-of-Experts (MoE) with word-specific experts acting as memory. MoWE surpasses T5 models in knowledge-heavy tasks while reducing computation (FLOPs), showcasing its efficiency in leveraging world knowledge. This technique optimizes language model performance by enhancing memory without inflating computational costs, marking a significant advancement in NLP tasks demanding extensive world knowledge incorporation.

The paper introduces LucidDreamer, a pioneering domain-free 3D scene generation pipeline leveraging large-scale diffusion models. By employing Dreaming and Alignment steps, it generates multi-view images using point clouds as guidance and integrates them into a unified 3D scene quality across diverse domains, supporting varied input types like text, RGB, and RGBD, allowing dynamic input changes during generation.

HierSpeech++ pioneers a fresh approach to speech synthesis, tackling the constraints of autoregressive models by unveiling a robust zero-shot speech synthesizer for both text-to-speech (TTS) and voice conversion (VC). Harnessing hierarchical speech synthesis frameworks, it notably enhances the resilience, richness, and speaker likeness of synthetic speech. Through its utilization of self-supervised speech representation and hierarchical variational auto encoders, this model attains a remarkable human-level quality in zero-shot speech synthesis, delivering an array of high-quality diverse outputs.

NeuroPrompts, a pioneering framework from Intel Labs, revolutionizes the text-to-image generation landscape. It ingeniously refines user prompts, optimizing image quality by aligning with expert prompt engineering styles. This adaptive system not only enhances image generation but also grants users the ability to finely tune stylistic features. NeuroPrompts’ code and a demo are accessible to the public, marking a significant leap forward in automated prompt optimization.

In this paper, the authors introduce a pioneering framework called DINOv that extends in-context prompting, commonly seen in large language models (LLMs), to improve vision tasks. While language-enhanced vision models have made strides, this work focuses on adapting an encoder-decoder architecture for visual in-context prompts. DINOv achieves impressive open-set segmentation tasks.

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.

Interested in reaching smart readers like you? To become an AI Breakfast sponsor, apply here.