💯 Guides 💯 

To start us off today, In my recent blog article, I explore Groq's innovative AI technology and delve into Groq's capabilities, demonstrating its groundbreaking chip architecture in action, designed to enhance AI inference speed, significantly outperforming existing standards. Follow along with full code samples to run the tests yourself and try out new examples. The results underscore Groq's superiority in speed with a 16x speed increase in my tests over GPT-4. This is a quick test but it easily shows why this is a technology to watch.

Read the full walkthrough on Bright Journey with AI

📰 News 📰 

Croissant Metadata Format for ML Datasets

Google Research introduces "Croissant," a new metadata format designed to streamline machine learning (ML) dataset organization and usage. Developed by the MLCommons community, Croissant enhances data structure and metadata layers, enabling easier dataset discovery, organization, and implementation in ML frameworks. Major ML platforms and tools like Kaggle, Hugging Face, and TensorFlow now support Croissant, facilitating the creation, search, and utilization of ML-ready datasets. This innovation aims to reduce the data preparation burden, making ML research and development more efficient. 

Read More At: Google AI Blog

OpenAI's Journey with Elon Musk: A Complex Relationship Explored

OpenAI details its history and evolving relationship with Elon Musk, focusing on the initial ambition to benefit humanity through AGI and the challenges of resource acquisition for this mission. Differences in vision led to Musk's departure, as he sought greater control while OpenAI prioritized mission integrity. Despite the split, OpenAI continues its commitment to making beneficial AI tools widely available, demonstrating real-world applications across different sectors. This insight into OpenAI’s perspective comes a few days after Elon Musk sued OpenAI for failing to uphold their mission of AGI for the good of humanity. The drama continues….

Read More At: OpenAI Blog

Salesforce Introduces Einstein 1 Studio for AI Customization

Salesforce announces the release of Einstein 1 Studio at the TrailblazerDX developer conference, enhancing generative AI capabilities for Salesforce developers. This platform enables customization of the Salesforce Einstein Copilot, facilitating the creation of specialized AI-driven actions and user experiences. Einstein 1 Studio includes tools like Copilot Builder, Prompt Builder, and Model Builder, catering to varying levels of coding expertise. The tools aim to simplify AI integration into business processes and extend the utility of Salesforce’s ecosystem through familiar developer tools like Apex and Heroku. Furthermore, the platform emphasizes flexibility in AI model integration, supporting a range of large language models (LLMs) and promoting customization without locking users into a single model.

Read More At: VentureBeat

UK Startup Launches AI Satellite for Real-Time Earth Images

UK-based startup Open Cosmos launched the AI-powered satellite HAMMER, aboard a SpaceX Falcon 9, aiming to provide near real-time Earth views, particularly of marine environments. The satellite, equipped with AI for quick data processing and a hyperspectral camera, aids in environmental monitoring and disaster response. This advancement enables instant data access via a mobile app and opens new possibilities for various applications including agriculture and energy monitoring. The launch aligns with Open Cosmos's expansion goals after raising $50 million.

Read More At: The Next Web

Ex-Google Engineer Accused of Stealing AI Secrets for China

Linwei Ding, a former Google software engineer, faces charges in the US for allegedly stealing AI-related trade secrets while working for two Chinese companies. Arrested in California, Ding is accused of taking over 500 confidential files from Google, especially concerning the tech giant's supercomputing data centers. He faces up to 10 years in prison and $250,000 in fines per charge if convicted. Ding, who joined Google in 2019, allegedly transferred these files to a personal account and was engaged with a Chinese startup, Beijing Rongshu Lianzhi Technology, and his own Shanghai-based AI firm, Shanghai Zhisuan Technology, without Google's knowledge. The case highlights the ongoing tensions between the US and China over technology and trade, with both nations striving for dominance in the AI field. 

Read More At: BBC News

Spain Halts Worldcoin Over Privacy Concerns

Spain has suspended the operations of Sam Altman's Worldcoin due to privacy concerns, specifically the technology's use of eyeball-scanning for digital ID and cryptocurrency distribution. Spanish regulators demand a halt to data collection and processing under EU's GDPR rules, highlighting issues like minor data collection and consent withdrawal. This is the first European action against Worldcoin, which faces scrutiny elsewhere too. The company asserts misunderstanding and seeks dialogue with regulators. 

Read More At: The Next Web

🧠 Research 🧠 

Design2Code: How Far Are We From Automating Front-End Engineering?

The paper explores the conversion of visual designs to code, focusing on real-world webpages. It introduces the Design2Code benchmark, a set of diverse real-world webpage examples, and automatic evaluation metrics. It evaluates current multimodal Large Language Models (LLMs), particularly GPT-4 Vision and Gemini Pro Vision, in converting screenshots to webpages. Findings show GPT-4 Vision outperforms others in most cases, with significant potential for replacing original webpages. The paper underscores the gap in transforming visual designs into functional code and suggests future research directions.

Read the full paper on ArXiv.

GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection

The authors introduce GaLore, a method enhancing the memory efficiency of Large Language Model (LLM) training. GaLore reduces memory demands by projecting gradient matrices into low-rank forms, enabling full-parameter learning while being more memory-efficient compared to existing techniques like Low-Rank Adaptation (LoRA). This approach allows for the pre-training of significant model architectures like LLaMA on standard GPUs without excessive memory requirements, fostering more accessible LLM training environments.

Read the full paper on ArXiv.

SaulLM-7B: A pioneering Large Language Model for Law

SaulLM-7B, a 7-billion-parameter large language model (LLM) specifically designed for legal text comprehension and generation, leveraging Mistral 7B architecture. It offers enhanced performance in legal tasks through a novel instructional fine-tuning method. SaulLM-7B aims to meet the increasing demand for AI in legal document analysis and interpretation. The model is open-source, licensed under the MIT License, promoting wider use and development in legal AI applications.

Read the full paper on ArXiv.

ShortGPT: Layers in Large Language Models are More Redundant Than You Expect

Xin Men et al. investigates the redundancy in large language models (LLMs) and proposes a novel pruning method based on a metric called Block Influence (BI). It demonstrates that significant redundancy exists at the layer level, allowing for the removal of layers without major loss in performance. The method, ShortGPT, significantly outperforms state-of-the-art methods in reducing parameters and computational needs while maintaining up to 92% performance.

Read the full paper on ArXiv.

🔨 AI Powered Tools 🔨 

A run down of the latest SaaS products and services which leverage AI to help you take back some time.

• Brevian - business users can build custom, secure AI agents that integrate with the team’s data and applications, answer queries and automate repeated tasks

• Socratic - Socratic was built to support Science, Math, Literature, Social Studies through visual explanations of important concepts.

• Khanmigo - If you have used Khan Academy (either yourself or with your kids) you’ll have to check out Khanmigo, your go-to AI tool for learning and teaching.

🗞️ Other News Rollup 🗞️ 

AI Safety Concerns - Microsoft engineer exposes Copilot's safety issues with objectification, violence, and copyrighted imagery, urging better oversight and reporting tools.

China's Tech Plans - China unveils 'AI Plus' initiative at National Congress to integrate AI into industry and enhance tech leadership.

AI Misleading Election Images - Generative AI tools create misleading election images despite safeguards. Platforms like Midjourney most likely to produce such content.

AI Partnership Boosts Enterprise Productivity - Accenture and Cohere collaborate to enhance AI accuracy and timeliness, revolutionizing the future of work with generative AI.

AI Risks in Science - AI poses long-term risks in scientific research by creating illusions of understanding. Crockett and Messeri caution against over-reliance.

Best AI Laptop - Apple's M3 MacBook Airs are hailed as the best consumer laptop for AI, offering powerful silicon and great design.

GPU Shipments Surge - Q4 2023 saw 32% growth in consumer-grade GPU shipments, despite AI PC hype. Prices stabilize and availability improves.

AI Grading Tools: Impact and Concerns - Educators use AI tool Writable for grading, raising concerns about outsourcing feedback. AI may save time but pose risks.

Robotics Expansion News - Hugging Face expands into robotics under ex-Tesla scientist, emphasizing open-source engineering in Paris.

🎶 Prompts 🎶 

Prompts used to generate some of this issues images. Unless otherwise stated all images are using Dall-E and ChatGPT Plus.

Thank You for Subscribing

Enjoying what you’re reading? Help me get better so I can continue to provide you with the most relevant content.