Daily Tech Digest - January 18, 2025

Beyond RAG: How cache-augmented generation reduces latency, complexity for smaller workloads

RAG is an effective method for handling open-domain questions and specialized tasks. It uses retrieval algorithms to gather documents that are relevant to the request and adds context to enable the LLM to craft more accurate responses. ... First, advanced caching techniques are making it faster and cheaper to process prompt templates. The premise of CAG is that the knowledge documents will be included in every prompt sent to the model. Therefore, you can compute the attention values of their tokens in advance instead of doing so when receiving requests. This upfront computation reduces the time it takes to process user requests. Leading LLM providers such as OpenAI, Anthropic and Google provide prompt caching features for the repetitive parts of your prompt, which can include the knowledge documents and instructions that you insert at the beginning of your prompt. ... And finally, advanced training methods are enabling models to do better retrieval, reasoning and question-answering on very long sequences. In the past year, researchers have developed several LLM benchmarks for long-sequence tasks, including BABILong, LongICLBench, and RULER. These benchmarks test LLMs on hard problems such as multiple retrieval and multi-hop question-answering. 


Turning Curiosity into a Career: The Power of OSINT

The beauty of OSINT is that you can start learning and practicing right now, even without a formal background in cybersecurity. Begin by familiarizing yourself with publicly available tools and resources. Social media platforms, search engines and public record databases are great starting points. From there, you can explore specialized tools like Google Dorking for advanced searches, reverse image search for photo analysis, and platforms like Maltego or SpiderFoot for more in-depth investigations. The OSINT Framework provides an extensive list of tools. If you're interested in pursuing OSINT as a career, consider taking advantage of free and paid online courses. Certifications such a GIAC Open Source Intelligence (GOSI) or Certified Ethical Hacker (CEH) can help build your credibility in the field. Participating in OSINT challenges or contributing to community projects is also a great way to hone your skills and showcase your abilities to potential employers. The demand for OSINT skills is growing as technology evolves and data becomes more accessible. Artificial intelligence and machine learning are enhancing OSINT capabilities, making it easier to analyze massive datasets and detect patterns. 


Five Trends That Will Drive Software Development in 2025

While organizations worldwide have quickly adopted AI for software development, many still struggle to measure its impact across diverse teams and business functions. Next year, organizations will become more sophisticated about measuring the return on their AI investments and better understand the value this technology can provide. This starts with looking more closely at specific outcomes. Instead of asking a broad question like, ‘How is AI helping my organization?’ leaders should study the impact of AI on tasks, such as test generation, documentation or language translation, and measure the gains in efficiency and productivity for these activities. ... While developers already work at breakneck speed today, technical debt is a persistent issue. The most worrying consequence of this debt is vulnerabilities that can creep into code and go unnoticed or unfixed. Next year, developers will expand their use of AI in software development to significantly reduce technical debt and increase the security of their code. Technical debt often occurs when developers choose an easy or quick solution instead of a better approach that takes longer. Vulnerabilities result when the code is poorly structured, not sufficiently reviewed or when testing is rushed or incomplete.


A Cloud Architect’s Guide to E-Commerce Data Storage

Latency, measured in microseconds, is the enemy of e-commerce storage systems, as slow-performing systems can mean hundreds of thousands of dollars in lost transactions and abandoned shopping carts. Your data platform must be reliable and highly performant even during fluctuating demand; events like Black Friday or unexpected social media trends can put a heavy load on your systems. Infrastructure that supports real-time data processing can be the deciding factor in staying competitive. These challenges necessitate a modern approach to storage — one that is software-defined, scalable and cloud-ready. ... Foundational elements of a modern e-commerce infrastructure consist of software-defined storage often combined with open-source environments like OpenStack, OpenShift, KVM and Kubernetes. The challenge for platform architects, whether building their e-commerce storage platform on premises or in the cloud, is to achieve scale and flexibility without compromising application and site performance. Many legacy storage systems, especially those architected for spinning disks, have performance limitations, resulting in data silos and expensive and time-consuming scaling strategies.


Demand and Supply Issues May Impact AI in 2025

Executives are asking for ROI numbers on analytics, data governance, and data quality programs, and they are demanding dollar values as opposed to “improving customer experience” or “increasing operational efficiency. ... Organizations have expected quick returns but not realized them because the initial expectations were unrealistic. Later comes the realization that the proper foundation has not been put in place. “Folks are saying they expect ROI in at least three years and more than 30% or so are saying that it would take three to five years when we’ve got two years of generative AI. [H]ow can you expect it to perform so quickly when you think it will take at least three years to realize the ROI? Some companies, some leadership, might be freaking out at this moment,” says Chaurasia. “I think the majority of them have spent half a million on generative AI in the last two years and haven’t gotten anything in return. That's where the panic is setting in.” Explaining ROI in terms of dollars is difficult, because it’s not as easy as multiplying time savings by individual salaries. Some companies are working to develop frameworks, however. ... If enterprises are reducing AI investments because the anticipated benefits aren’t being realized, vendors will pull back. 


4 Strategies To Thrive In A Manager-Less Workplace

One of the most important skills you can build is emotional regulation. Work can be intense, often frustrating. It’s easy to get caught up in your own emotions and—since emotions are catching—other people’s as well. Staying even-keeled pays off in maintaining good relationships with peers and also keeping yourself clear-headed so you can problem-solve when things go wrong. You can work on your emotional self-control by learning the tools of journaling and mindfulness. ... When you communicate powerfully, you navigate more easily. You get what you need more efficiently, you sell your ideas, and you build better relationships. All of these outcomes are useful when you’re on your own to build a case for getting promoted. The best way to build these skills is to practice. Volunteer to give large presentations and ask for feedback. Craft your emails and slack messages with an understanding of the receiver and ask them if they have suggestions for you. ... Your network inside your company can also provide the emotional support you would have gotten from your manager. And, when it comes time for you to be promoted, in most companies you need your colleagues to support you. Look around at your coworkers to see who are the most interesting, plugged-in, or effective. 


Dark Data: Recovering the Lost Opportunities

Dark data is the data collected and stored by an organization but is not analyzed or used for any essential purpose. It is frequently referred to as "data that lies in the shadows" because it is not actively used or essential in decision-making processes. ... Dark data can be highly beneficial to businesses as it offers insights and business intelligence that wouldn't be available otherwise. Companies that analyze dark data can better understand their customers, operations, and market trends. This enables them to make the best decisions and improve overall performance. Dark data can help organizations recoup lost opportunities by uncovering previously unknown patterns and trends. ... Once the dark data has been collected, it must be cleansed before further analysis. This may include deleting duplicate data, correcting errors, and formatting information to make it easier to work with. After the data has been cleansed and categorized, it can be examined to reveal patterns and insights that will aid decision-making. ... Collaborating with cross-functional teams, such as IT, data science, and business divisions, can assist in guaranteeing that dark data is studied in light of the organization's broader goals and objectives. 
The difference between “data deletion” and “data destruction” is critical to understand. “Data deletion” simply means removing a file from a system, making it appear inaccessible, while “data destruction” is a more thorough process that permanently erases data from a storage device, making it completely irretrievable. Deleting data isn’t enough. Without proper destruction protocols, “deleted” data remains vulnerable to breaches, regulatory compliance, and data recovery tools. ... A well-defined data destruction policy is your organization’s first line of defense. It outlines when, how, and under what circumstances data should be destroyed. Without a formal policy, data is often overlooked, forgotten, or destroyed haphazardly, creating compliance and security risks. To implement this, start by identifying the types of data your organization collects and classifies, such as PII or proprietary records. Define clear retention periods based on regulatory requirements like GDPR or CCPA and document the necessary steps, tools, and roles for secure destruction. Assign accountability to ensure oversight and follow-through. A formal policy isn’t just a “nice-to-have.” It’s a compliance requirement for many regulations, including GDPR and CCPA. 


Can GenAI Restore the ‘Humanity’ in Banking that Digital Has Removed?

Abbott is not arguing for turning customers directly over to GenAI — not yet. Even the most-advanced pioneers his firm works with aren’t risking that. ... Abbott believes GenAI, as it becomes a standard part of banking, will play out in a similar way. Employees will adapt, often more slowly than anticipated, but they will change. This will lead to shifts in the role of management vis-à-vis employees empowered by GenAI. Abbott says this will likely take a similar path to that seen as banks adopted agile development. Young people came into the bank using the tools, just as many are already experimenting with GenAI. Banking leaders liked the idea of their organizations "doing agile." But what Abbott calls "the frozen middle" management tier had to grin and plunge into unfamiliar turf. "That frozen middle will have to thaw out and find a new way of working," says Abbott. Bank leadership must help by providing tools and opportunities for trying it out. One of the biggest early challenges will be tempering the GenAI tech to the task. Abbott explains that GenAI can be tuned to be "low temperature" or "high temperature," or somewhere in between. The former refers to GenAI working with tight guardrails, such as in sensitive areas like dispute management. 


Federated learning: The killer use case for generative AI

Federated learning is emerging as a game-changing approach for enterprises looking to leverage the power of LLMs while maintaining data privacy and security. Rather than moving sensitive data to LLM providers or building isolated small language models (SLMs), federated learning enables organizations to train LLMs using their private data where it resides. Everyone who worries about moving private enterprise data to a public space, such as uploading it to an LLM, can continue to have “private data.” Private data may exist on a public cloud provider or in your data center. The real power of federation comes from the tight integration between private enterprise data and sophisticated LLM capabilities. This integration allows companies to leverage their proprietary information and broader knowledge in models like GPT-4 or Google Gemini without compromising security. ... As enterprises struggle to balance AI capabilities against data privacy concerns, federated learning provides the best of both worlds. Also, it allows for a choice of LLMs. You can leverage LLMs that are not a current part of your ecosystem but may be a better fit for your specific application. For instance, LLMs that focus on specific verticals are becoming more popular. 



Quote for the day:

"Too many of us are not living our dreams because we are living our fears." -- Les Brown

No comments:

Post a Comment