Weekly Bytes about AI, Data Science & Web3 - CW 51

This Week in AI, Data Science and Web3

Welcome to our latest newsletter, packed with the latest insights and advances in AI, Web3 and data science. Dive into the critical role of AI governance for corporate boards and explore Google DeepMind's breakthrough in depth estimation with the DMD model. Discover tldraw's innovative AI-driven prototyping tool and IBM's strategic move to bolster its hybrid cloud infrastructure with a $2.3 billion acquisition. In Web3, we delve into developments in Ethereum's latency optimization and Revolut's response to FCA regulations. Data science enthusiasts can look forward to insights on Qlik's latest acquisition and the challenges of scaling generative AI, as well as evaluations of large language models.

?? Artificial Intelligence

Strategic Oversight: Boards' Role in AI Governance

As artificial intelligence becomes a critical factor in business strategy, boards of directors must navigate its complexities. Dr. Benjamin van Giffen and Helmuth Ludwig from the Universit?t St.Gallen (HSG) and Southern Methodist University outline key focus areas for board-level AI governance: enhancing strategy and competitiveness, judicious capital allocation, understanding AI-specific risks, and fostering technology competence. The study stresses that effective AI governance requires boards to tailor their approach to their organization's unique context and demands. It offers practical guidance for boards to evaluate their AI readiness and drive strategic discussions. More on this.

Advancing Zero-Shot Metric Depth Estimation: Google's DMD Model

Google DeepMind 's latest innovation, the Diffusion for Metric Depth (DMD) model, represents a significant breakthrough in zero-shot metric depth estimation, particularly for monocular depth estimation. This advancement addresses the complexities of jointly modeling indoor and outdoor scenes, traditionally challenged by varying RGB and depth distributions and depth-scale ambiguity due to unknown camera intrinsics. DMD diverges from previous specialized multi-head architectures, adopting a task-agnostic diffusion model with log-scale depth parameterization. It incorporates field-of-view (FOV) conditioning to manage scale ambiguity and leverages synthetic FOV augmentation during training. This approach enables DMD to transcend the limitations of camera intrinsics in training datasets. Notably, DMD achieves a remarkable 25% reduction in relative error on zero-shot indoor and 33% reduction on zero-shot outdoor datasets compared to the current state-of-the-art, using only a minimal number of denoising steps. More on this.

tldraw's AI-Powered Web Prototyping Tool

Peter Gostev , recently shared his enthusiasm for tldraw , an innovative AI tool that transforms sketched designs into functional HTML pages. With its 'Make it Real' feature, tldraw enables users to quickly prototype web pages on an unlimited whiteboard. The addition of the 'Open in Replit' button further enhances this tool by allowing users to edit and deploy their apps in a web-based IDE. Gostev demonstrated this capability by creating a web app prototype for the Pro Prompter idea, showcasing the ease and speed of developing from a simple sketch to a deployed prototype. The entire process, from sketching to deployment, is streamlined and accessible, even for those with limited coding experience. The deployed prototype and the project on Replit are available for exploration through following links:

Here's the deployed prototype: https://lnkd.in/ekeMzF57

Here's the Replit project generated: https://lnkd.in/erBair-R

Users interested in trying tldraw need their own OpenAI API Key.

IBM Acquires StreamSets and WebMethods for $2.3 Billion

IBM has announced a significant €2.13 billion ($2.3 billion) acquisition of data integration assets StreamSets and WebMethods from Software AG , a move that aligns with IBM's strategy towards hybrid cloud computing. StreamSets, acquired by Software AG last year, and WebMethods, purchased in 2007, bolster IBM's integration platform-as-a-service (IPAAS) offerings. This acquisition is part of IBM's larger push into hybrid cloud infrastructure, following its massive purchases like the $34 billion Red Hat acquisition in 2018 and the recent $4.6 billion Apptio deal. These technologies are crucial for businesses operating across various cloud and on-prem environments, facilitating seamless data management and integration. Additionally, the integration of these tools with IBM’s Watsonx AI and data platform will enhance the company's capabilities in AI, application modernization, and data management. More on this.

?? Web3

Chorus One's Adagio Study on MEV and Latency Optimization

Chorus One 's empirical study, "Adagio," explores the effects of latency optimization on Maximal Extractable Value (MEV) in Ethereum . The study reveals that latency optimization significantly boosts MEV capture, with an Annual Percentage Rate (APR) increase of 1.58% compared to standard setups. However, this increase in MEV rewards comes at a cost, potentially leading to higher gas fees and increased centralization risks. The findings also suggest that larger node operators, with more voting power, benefit more from MEV, further intensifying the push towards strategic latency optimization among all node operators. The study emphasizes the need for a balanced approach in employing latency tactics, considering their broader impact on the Ethereum network. More on this.

Coinbase Expands into France with New Crypto License Amid U.S. Regulatory Challenges

Coinbase , a leading cryptocurrency exchange, has gained a significant foothold in Europe with its recent registration approval by France's AMF watchdog. This Virtual Asset Service Provider (VASP) approval allows Coinbase to operate various digital currency services in France, including digital asset custody and trading. This move is part of France's ambition, under President Emmanuel Macron, to become a global tech hub, investing heavily in technologies like AI and crypto. The European Union's upcoming Markets in Crypto Assets (MiCA) regulation will further streamline crypto operations across the bloc, enabling companies like Coinbase to expand more easily. This expansion into Europe comes at a critical time for Coinbase, as it faces a more uncertain regulatory environment in the United States, where the SEC has been aggressively targeting crypto firms. France's growing crypto adoption and its positioning as a technology leader make it an attractive market for crypto companies. More on this.

Bitcoin Fees Rise Due to Ordinals, But It May Benefit the Network

The Bitcoin blockchain is experiencing increased congestion and rising transaction fees, primarily due to the introduction of Ordinal inscriptions. Ordinals, which embed unique data such as art onto the blockchain, have been a contentious topic in the Bitcoin community. Traditionalists argue that Ordinals deviate from Bitcoin's original purpose of peer-to-peer financial transactions, while supporters see them as beneficial for the network's long-term security. They argue that higher transaction fees incentivize miners, ensuring network health. Despite concerns, the popularity of Ordinals is undeniable, with significant sales of Bitcoin-based digital art. The discussion echoes previous debates over the incorporation of unique digital assets onto the Bitcoin blockchain, dating back to the OP_RETURN function in 2013 and the Counterparty protocol in 2014. This current trend of Ordinals highlights the ongoing evolution of Bitcoin's usage and the complex balancing act between innovation and maintaining the network's foundational principles. More on this.

In addition, in a significant milestone for the Bitcoin network, the number of Bitcoin addresses with non-zero balances has surpassed 50 million, according to Glassnode . This achievement underscores the growing adoption and use of bitcoin around the world.

Ledger's Response to Security Breach: Ensuring User Safety and Future Standards

Ledger has taken decisive action following a recent security breach, committing to fully compensate the approximately $600k stolen from users during a blind signing incident on EVM DApps. By June 2024, Ledger will eliminate the option of blind signing on its devices, shifting towards Clear Signing to allow users to verify transaction details on a secure display. This change aims to set a new standard in DApp security and protect users from similar front-end attacks. Ledger reassures that its devices and Ledger Live were not compromised and remains in contact with affected users, ensuring resolution by February 2024. They also encourage DApp developers to collaborate in adopting Clear Signing, enhancing overall ecosystem security. More on this.

Revolut Pauses UK Business Crypto Purchases Due to FCA Regulations

Revolut is temporarily suspending cryptocurrency purchases for its UK Business customers from January 3rd, in response to new rules from the Financial Conduct Authority (FCA) effective January 8th. This pause, however, will not affect other Revolut customers. The FCA's new requirements are aimed at enhancing customer experiences and providing additional protection for crypto asset investors. Revolut Business clients can still hold or sell existing crypto assets, and the company will notify them when the suspension is lifted. This move follows the FCA's stricter regulations rolled out earlier this year, focusing on clearer, fairer marketing and risk warning requirements for crypto products.

Wynd Network Secures $3.5M for Decentralized AI Initiative

Wynd Labs , developing the Grass app for decentralized web scraping, has raised $3.5 million in a seed round led by Polychain Capital and Tribe Capital . Grass, a browser extension, utilizes users' unused internet bandwidth to scrape web data and rewards them with Grass Points. This data is then sold to AI developers. Wynd's approach, leveraging blockchain, ensures unbiased data training for AI models. The company is also exploring possibilities for a future token and is working with open-source AI firms, highlighting the growing interest in integrating AI and blockchain technology. More on this.

Dune Unveils Exciting New Features for 2024

Dune Analytics ushers in 2024 with a suite of new features aimed at enhancing the data analytics experience. The updates include the general availability of Alerts, allowing users to receive notifications for scheduled queries through email or webhooks. Significant price reductions have been implemented for datapoints in Plus and Premium plans, with increased datapoints per credit and streamlined CSV downloads. CRUD query endpoints are now accessible in all paid plans, enabling programmatic query management. A much-requested Dark Mode offers a visually soothing alternative interface, ideal for late-night sessions. The introduction of Command K enhances keyboard navigation efficiency across the platform. Additionally, the new chart screenshot feature facilitates easy sharing of insights, complete with a handy camera icon for quick access. More on this.

Guidance on Staking Governance and Legal Implications in Switzerland

The Eidgen?ssische Finanzmarktaufsicht FINMA has released guidance on staking, particularly regarding its supervisory treatment. With the DLT Act providing a basis for crypto asset custody, FINMA acknowledges the complexity of staking services and their varied implications. The guidance delineates custodial staking, which includes direct staking and staking chains, as well as non-custodial staking where customers retain control over withdrawal keys. Risks associated with staking such as technical malfunctions, slashing due to node misbehavior, counterparty risks in bankruptcy, and market risks during volatile periods are highlighted. FINMA stresses the importance of assets being readily available for customers to qualify for bankruptcy protection. The guidance also addresses the legal ambiguities surrounding staking, especially when lock-up periods or slashing are involved, and outlines interim practices for licensed institutions until legislation or court decisions provide further clarity. More on this.

PwC's Global Crypto Landscape 2024 Outlook

普华永道 's second annual Global Crypto Landscape report provides a comprehensive overview of the evolving global regulatory landscape for digital assets. Matt Blumenfeld , Web3 & Digital Asset Lead at PwC US, emphasizes that 2024 is pivotal for the industry, not just for surviving challenges but for building a stable ecosystem with clear regulatory guidelines as its foundation. The report discusses how different jurisdictions are defining digital assets, with terms ranging from "crypto asset" to "virtual currency." It explores the role of jurisdictional equivalence in global crypto regulations, the implications of choosing a jurisdiction for incorporation and operations, and recent developments by global standard-setting institutions. Significant focus is given to the EU's advancements in digital asset regulation, including the MiCAR, and jurisdictional-level regulatory summaries globally. This report serves as a key resource for understanding the impact of regulatory frameworks on both crypto and traditional financial services firms. More on this.

?? Data Science

Qlik Acquires Mozaic Data to Revolutionize Data Management

Qlik Technologies Inc. has significantly advanced its data management capabilities with the acquisition of AI-driven technology company Mozaic Data. This strategic move, announced on December 19, 2023, aims to simplify enterprise data handling, emphasizing a product-oriented approach for enhanced data quality and governance. Integrating Mozaic's innovative technology, Qlik is set to offer a transformative Data Product Catalog experience, leveraging Sharad Kumar's expertise, Mozaic's founder, now part of Qlik. This acquisition aligns with the growing need for efficient, AI-driven data management solutions, promising to reshape data consumption and deployment across various cloud platforms like Amazon Redshift and Google BigQuery. With Mozaic's integration, Qlik strengthens its position in providing comprehensive, end-to-end data solutions for businesses. More on this.

Forrester Study Highlights Need for Capable Platforms in Scaling Generative AI

The Forrester Opportunity Snapshot, commissioned by Dataiku , reveals a strong inclination towards Generative AI (GenAI) among large North American companies. With 83% of respondents actively exploring GenAI, the focus is on enhancing customer experiences and developing new products or services. However, organizations face challenges in operationalizing GenAI, particularly in governance and technical infrastructure. The study emphasizes the need for end-to-end AI platforms that facilitate integration, governance, and scalability. Dataiku emerges as a leader in this space, providing a structured environment for seamless integration and collaborative capabilities. This platform approach is seen as essential for effectively implementing and scaling GenAI solutions, addressing infrastructure challenges and ensuring smooth incorporation into organizational processes. More on this.

Deepset Explores Evaluating Large Language Models in RAG Pipelines

deepset presents an informative exploration of the evaluation of Large Language Models (LLMs) within Retrieval Augmented Generation (RAG) pipelines, a crucial component in the field of language modeling. This second article in the mini-series examines the role of the LLM in a RAG pipeline and presents various methods for LLM evaluation, including lexical and semantic metrics. It also emphasizes the irreplaceable value of user feedback in assessing the real-world utility of these models. This comprehensive article serves as a guide on how to start evaluating LLMs, highlighting the unique challenges and promising solutions in the field of generative AI. More on this.

Supabase x OpenAI Cookbook: A Guide to Semantic Search with Vector Databases

Paul Copplestone , CEO of Supabase , announced the release of the Supabase x OpenAI Cookbook. This comprehensive guide offers step-by-step instructions on setting up a vector database using pgvector, a PostgreSQL extension. It demonstrates how to create a SQL table capable of storing vector data and how to generate OpenAI embeddings using OpenAI's JavaScript client. The guide further explains storing these embeddings in the SQL table and conducting semantic searches over them using a PostgreSQL function and the Supabase JavaScript client. This resource is ideal for developers looking to integrate advanced AI services and tools. More on this.

Data Management For Large Language Models

The School of Computer Science at Peking University have released an insightful study on the significance of data management in the training of Large Language Models (LLMs). It highlights the critical role of data handling in both the pretraining and supervised fine-tuning stages to improve the performance and efficiency of LLMs. It discusses pivotal aspects like data quantity, quality, domain composition, and data management systems for pretraining, as well as data quality, task composition, and data-efficient learning strategies for fine-tuning. The study addresses key challenges such as mitigating social biases, handling hallucinations in model outputs, and the absence of a unified data management framework across various applications. This research serves as a valuable guide for practitioners in the field and sets the stage for future innovations in overcoming the existing hurdles in data management for LLM development. More on this.

LlamaIndex Partners with Activeloop for Free Course in RAG

LlamaIndex , in partnership with Activeloop , has launched a comprehensive and free course dedicated to retrieval-augmented generation for production. This educational initiative includes 33 lessons and 7 practical assignments, culminating in a certification for participants. Designed to provide an extensive learning experience, the course offers over 25 hours of content, completely free of charge. This program is an excellent opportunity for individuals interested in deepening their understanding and skills in retrieval-augmented generation, a pivotal area in AI and machine learning. Enrollment is open, and interested learners can sign up through this link.

?? Special this Week

Ezra Eeman has shared the Wayfinder Review/Preview Report for 23-24, which offers a comprehensive analysis of the past year's digital and media trends, and provides foresight into the upcoming developments. The report serves as a guide to navigate the rapidly evolving digital landscape, focusing on AI advancements, community engagement, and new consumption patterns within creative industries and digital spaces. ?? https://lnkd.in/e6Vmg6Jg

?? Merry X-Mas

As we close this newsletter, we would like to wish all of our readers a very Merry Christmas. May this festive season bring joy, peace and happiness to you and your loved ones. Let's cherish these moments of togetherness and look forward to a new year filled with new opportunities and advancements in the world of AI, Data Science and Web3. Merry Christmas and Happy Holidays! ??

Hey there, we are Sandro and Fabian!

Passionate about ???????? and Blockchain? So are we.

To stay in the loop:

?? Smash the "Subscribe" button for Weekly Bytes

?? Ring the ?? on our profile

?? Let's connect!

Disclaimer: The following information is provided for educational and informational purposes only. It should not be construed as financial or investment advice. Investing in cryptocurrencies, NFTs, or any other asset class involves inherent risks, and individuals should conduct their own research (DYOR — Do Your Own Research) and seek professional advice before making any investment decisions. The content provided does not guarantee any specific outcomes or results. The responsibility for any investment or financial decisions rests solely with the individual.