Daily Tech Digest - August 26, 2024

The definitive guide to data pipelines

A key data pipeline capability is to track data lineage, including methodologies and tools that expose data’s life cycle and help answer questions about who, when, where, why, and how data changes. Data pipelines transform data, which is part of the data lineage’s scope, and tracking data changes is crucial in regulated industries or when human safety is a consideration. ... Other data catalog, data governance, and AI governance platforms may also have data lineage capabilities. “Business and technical stakeholders must equally understand how data flows, transforms, and is used across sources with end-to-end lineage for deeper impact analysis, improved regulatory compliance, and more trusted analytics,” says Felix Van de Maele, CEO of Collibra. The data ops behind data pipelines When you deploy pipelines, how do you know whether they receive, transform, and send data accurately? Are data errors captured, and do single-record data issues halt the pipeline? Are the pipelines performing consistently, especially under heavy load? Are transformations idempotent, or are they streaming duplicate records when data sources have transmission errors?


Living with trust issues: The human side of zero trust architecture

As we’ve become more dependent on technology, IT environments have become more complex. This has made threats more intense and could even pose a serious danger. To tackle these growing security challenges — which needed a stronger and more flexible approach — industry experts, security practitioners, and tech providers came together to develop the zero trust architecture (ZTA) framework. This development led to a growing recognition of the importance of prioritizing verification over trust, which made ZTA a cornerstone of modern cybersecurity strategies. The main idea behind ZTA is to “never trust, always verify.” ... Implementing the ZTA framework means that every action the IT and security teams handle is filtered through a security-first lens. However, the over-repeated mantra of “never trust, always verify” may affect the psychological well-being of those implementing it. Imagine spending hours monitoring every network activity while constantly questioning if the information is genuine and if people’s motives are pure. This suspicious climate not only affects the work environment but also spills over into personal interactions, affecting trust with others. 


Top technologies that will disrupt business in 2025

Chaplin finds ML useful for identifying customer-related trends and predicting outcomes. That sort of forecasting can help allocate resources more effectively, he says, and engage customers better — for example when recommending products. “While gen AI undoubtedly has its allure, it’s important for business leaders to appreciate the broader and more versatile applications of traditional ML,” he says. ... What Skillington touches on is the often-overlooked facet of any successful digital transformation: It all starts with data. By breaking down data silos, establishing wholistic data governance strategies, developing the right data architecture for the business, and developing data literacy across disciplines, organizations can not only gain better access to their data but also better understand how ... Edge computing and 5G are two complementary technologies that are maturing, getting smaller, and delivering tangible business results securely, says Rogers Jeffrey Leo John, CTO and co-founder of DataChat. “Edge devices such as mobile phones can now run intensive tasks like AI and ML, which were once only possible in data centers,” he says. 


Meta presents Transfusion: A Recipe for Training a Multi-Modal Model Over Discrete and Continuous Data

Transfusion is trained on a balanced mixture of text and image data, with each modality being processed through its specific objective: next-token prediction for text and diffusion for images. The model’s architecture consists of a transformer with modality-specific components, where text is tokenized into discrete sequences and images are encoded as latent patches using a variational autoencoder (VAE). The model employs causal attention for text tokens and bidirectional attention for image patches, ensuring that both modalities are processed effectively. Training is conducted on a large-scale dataset consisting of 2 trillion tokens, including 1 trillion text tokens and 692 million images, each represented by a sequence of patch vectors. The use of U-Net down and up blocks for image encoding and decoding further enhances the model’s efficiency, particularly when compressing images into patches. Transfusion demonstrates superior performance across several benchmarks, particularly in tasks involving text-to-image and image-to-text generation. 


AI Assistants: Picking the Right Copilot

The best assistant operates as an agent that understands what context the underlying AI can assume from its known environment. IDE assistants such as GitHub Copilot know that they are responding with programming projects in mind. GitHub Copilot examines script comments as well as syntax in a given script before crafting a suggestion. The tool examines syntax and comments against its trained datasets, consisting of GPT training and the codebase of GitHub's public repositories. GitHub Copilot was trained on the public repositories in GitHub, so it has a slightly different "perspective" on syntax than that of ChatGPT ADA. Thus, the choice of corpus for an AI model can influence what answer an AI assistant yields to users. A good AI assistant should offer a responsive chat feature to indicate its understanding of its environment. Jupyter, Tabnine, and Copilot all offer a native chat UI for the user. The chat experience influences how well a professional feels the AI assistant is working. How well it interprets prompts and how accurate the suggestions are all start with the conversational assistant experience, so technical professionals should note their experiences to see which assistant works best for their projects.


Is the vulnerability disclosure process glitched? How CISOs are being left in the dark

The elephant in the room regarding misaligned motives and communications between researchers and software vendors is that vendors frequently try to hide or downplay the bugs that researchers feel obligated to make public. “The root cause is a deep-seated fear and prioritizing reputation over security of users and customers,” Rapid7’s Condon says. “What it comes down to many times is that organizations are afraid to publish vulnerability information because of what it might mean for them legally, reputationally, and financially if their customers leave. Without a concerted effort to normalize vulnerability disclosure to reward and incentivize well-coordinated vulnerability disclosure, we can pick at communication all we want. Still, the root cause is this fear and the conflict that it engenders between researchers and vendors.” Condon is, however, sympathetic to the vendors’ fears. “They don’t want any information out there because they are understandably concerned about reputational damage. They’re seeing major cyberattacks in the news, CISOs and CEOs dragged in front of Congress or the Senate here in the US, and lawsuits are coming out against them. ...”


Level Up Your Software Quality With Static Code Analysis

Behind high-quality software is high-quality code. The same core coding principles remain true regardless of how the code was written, either by humans or AI coding assistants. Code must be easy to read, maintain, understand and change. Code structure and consistency should be robust and secure to ensure the application performs well. Code devoid of issues helps you attain the most value from your software. ... While static analysis focuses on code quality and reduces the number of problems to be found later in the testing stage, application testing ensures that your software actually runs as it was designed. By incorporating both automated testing and static analysis, developers can manage code quality through every stage of the development process, quickly find and fix issues and improve the overall reliability of their software. A combination of both is vital to software development. In fact, a good static analysis tool can even be integrated into your testing tools to track and report the percentage of code covered by your unit tests. Sonar recommends a test code coverage of 80% or your code will fail to pass the recommended standard.


Two strategies to protect your business from the next large-scale tech failure

The key to mitigating another large-scale system failure is to plan for catastrophic events and practice your response. Make dealing with failure part of normal business practices. When failure is unexpected and rare, the processes to deal with it are untested and may even result in actions which make the failure worse. Build a network and a team that can adapt and react to failures. Remember when insurance companies ran their own data centres and disaster recovery tests were conducted twice a year? ... The second strategy for minimizing large-scale failures is to avoid the software monoculture created by the concentration of digital tech suppliers. It’s more complex but worth it. Some corporations have a policy of buying their core networking equipment from three or four different vendors. Yes, it makes day-to-day management a little more difficult, but they have the assurance that if one vendor has a failure, their entire network is not toast. Whether it’s tech or biology, a monoculture is extremely vulnerable to epidemics which can destroy the entire system. In the CrowdStrike scenario, if corporate networks had been a mix of Windows, Linux and other operating systems, the damage would not have been as widespread.


India's Critical Infrastructure Suffers Spike in Cyberattacks

The adoption of emerging technologies such as AI and cloud and the focus on innovation and remote working has driven digital transformations, thus boosting companies' need for more security defenses, according to Manu Dwivedi, partner and leader for cybersecurity at consultancy PwC India. "AI-enabled phishing and aggressive social engineering have elevated ransomware to the top concern," he says. "While cloud-related threats are concerning, greater interconnectivity between IT and OT environments and increased usage of open-source components in software are increasing the available threat surface for attackers to exploit." Indian organizations also need to harden their systems against insider threats, which requires a combination of business strategy, culture, training, and governance processes, Dwivedi says. ... The growing demand for AI has also shaped the threat landscape in the country and threat actors have already started experimenting with different AI models and techniques, says PwC India's Dwivedi. "Threat actors are expected to use AI to generate customized and polymorphic malware based on system exploits, which escapes detection from signature-based and traditional detection methods," he says.


Architectural Patterns for Enterprise Generative AI Apps

In the RAG pattern, we integrate a vector database that can store and index embeddings (numerical representations of digital content). We use various search algorithms like HNSW or IVF to retrieve the top k results, which are then used as the input context. The search is performed by converting the user's query into embeddings. The top k results are added to a well-constructed prompt, which guides the LLM on what to generate and the steps it should follow, as well as what context or data it should consider. ... GraphRAG is an advanced RAG approach that uses a graph database to retrieve information for specific tasks. Unlike traditional relational databases that store structured data in tables with rows and columns, graph databases use nodes, edges, and properties to represent and store data. This method provides a more intuitive and efficient way to model, view, and query complex systems. ... Like the basic RAG system, GraphRAG also uses a specialized database to store the knowledge data it generates with the help of an LLM. However, generating the knowledge graph is more costly compared to generating embeddings and storing them in a vector database. 



Quote for the day:

"Leadership is a matter of having people look at you and gain confidence, seeing how you react. If you're in control, they're in control." -- Tom Landry

No comments:

Post a Comment