Daily Tech Digest - July 25, 2024

7 LLM Risks and API Management Strategies to Keep Data Safe

Overloading an LLM with requests can cause poor service or increased resource costs, two of the worst outcomes for an organization. Yet with a model denial of service that is what’s at stake. This happens when attackers cause resource-heavy operations on LLMs. This could look like a higher-than-normal task generation or repeated long inputs, to name a few. Authentication and authorization can be used to prevent unauthorized users from interacting with the LLM. Rate limiting on the number of tokens per user should also be used to stop users from burning through an organization’s credits, incurring high costs and using large amounts of computation resulting in latency injection. ... Compliance teams’ concern about sensitive information disclosure is perhaps one of the most severe vulnerabilities limiting LLM adoption. This occurs when models inadvertently can return sensitive information, resulting in unauthorized data access, privacy violations and security breaches. One technique that developers can implement is using specially trained LLM services to identify and either remove or obfuscate sensitive data.


Michael Dell performed a ‘hard reset’ of his company so it could survive massive industry shifts and thrive again. Here’s how it’s done

A hard reset asks and answers a small set of critical strategy questions. It starts with revisiting your beliefs. Discuss and debate your updated beliefs with the team and build a plan to actively test the ones where you disagree or have the most uncertainty about. Next ask what it will take to build a defensible competitive advantage going forward: Determine if you still have a competitive advantage (you probably don’t—otherwise you wouldn’t be in a hard reset). Glean what elements you can use to strengthen and build an advantage going forward. Over-index on the assets you can strengthen and discuss what you will buy or build. Make sure you anchor this in your beliefs around where the world is going. ... During a hard reset, develop rolling three-month milestones set towards a six-month definition of success. Limit these milestones to ten or fewer focused tasks. Remember you are executing these milestones while continuing the reset process and related discussions, so be realistic with what you can achieve and avoid including mere operational tactics on the milestone list.


Software testing’s chaotic conundrum: Navigating the Three-Body Problem of speed, quality, and cost

Companies that prioritize speed over quality end up with the choice of whether to release to market anyway, and risk reputational damage and client churn, or push back timelines and go over budget trying to retrofit quality (which isn’t really possible, by the way). ... Quality is the cornerstone of successful digital products. Users expect software to function reliably, deliver on its promises and provide a seamless user experience. Comprehensive testing plays a large role in making sure users are not disappointed. Developers need to look beyond basic functional testing and consider aspects like accessibility, payments, localisation, UX and customer journey testing. However, investing heavily in testing infrastructure, employing skilled QA engineers and rigorously testing every feature before release is expensive and slow. ... Quality engineers are limited by budget constraints, which can affect everything from resource allocation to investments in tooling. However, underfunding quality efforts can have disastrous effects on customer satisfaction, revenues and corporate reputation. To deliver competitive products within a reasonable timeframe, quality managers need to use available budgets as efficiently as possible. 


Cloud security threats CISOs need to know about

An effective cloud security incident response plan details preparation, detection and analysis, containment, eradication, recovery and post-incident activities. Preparation involves establishing an incident response team with defined roles, documented policies, necessary tools and a communication plan for stakeholders. Detection and analysis require continuous monitoring, logging, threat intelligence, incident classification and forensic analysis capabilities. Containment strategies and eradication processes are essential to prevent the spread of incidents and eliminate threats, followed by detailed recovery plans to restore normal operations. Post-incident activities include documenting actions, conducting root cause analysis, reviewing lessons learned, and updating policies and procedures. ... Organizations should start by doing a comprehensive risk assessment to identify critical assets and evaluate potential risks, such as natural disasters and cyberattacks. Following the assessment, develop and document DR and BC procedures. Annually review and update the procedures to reflect changes in the IT environment and emerging threats.


Artificial Intelligence Versus the Data Engineer

So, how does AI change the role of the data engineer? Firstly, the role of the data engineer has always been tricky to define. We sit atop a large pile of technology, most of which we didn’t choose or build, and an even larger pile of data we didn’t create, and we have to make sense of the world. Ostensibly, we are trying to get to something scientific. A number, a chart, a result that we can stand behind and defend—but like all great science, getting there also needs a bit of art. That art comes in the form of the intuition required to sift through the data, understand the technology, and rediscover all the little real-world nuances and history that over time have turned some lovely clean data into a messy representation of the real world. ... What’s exciting for us beleaguered data engineers is that AI is showing great ability to be a very helpful tool for these hard-to-master skills that will ultimately make us better and more productive at our jobs. We have all, no doubt, seen all the great advancements in AI’s ability to take plain text queries and turn them into increasingly complex SQL, thus lightening the load of remembering all the advanced syntax for whichever data platform is in vogue.


CrowdStrike crash showed us how invasive cyber security software is. Is there a better way?

In the wake of this incident it’s worth considering whether the tradeoffs made by current EDR technology are the right ones. Abandoning EDR would be a gift to cyber criminals. But cyber security technology can – and should – be done much better. From a technical standpoint, Microsoft and CrowdStrike should work together to ensure tools like Falcon operate at arm’s length from the core of Microsoft Windows. That would greatly reduce the risk posed by future faulty updates. Some mechanisms already exist that may allow this. Competing technology to CrowdStrike’s Falcon already works this way. To protect user privacy, EDR solutions should adopt privacy-preserving methods for data collection and analysis. Apple has shown how data can be collected at scale from iPhones without invading user privacy. To apply such methods to EDR, though, we’ll likely need new research. More fundamentally, this incident raises questions about why society continues to rely on computer software that is so demonstrably unreliable. 


6 Pillars Of Entrepreneurial Mastery: Elevating Your Business Through Lifelong Learning

Entrepreneurs with a growth mindset understand that abilities and intelligence can be developed through dedication and hard work. This perspective fosters resilience, helping to navigate setbacks and failures with a constructive attitude. By viewing challenges as opportunities for growth, you can become more adaptable and willing to take calculated risks. Regular self-reflection, seeking feedback and staying open to new ideas are essential practices for cultivating this mindset. ... As an entrepreneur, continuously educate yourself on tax regulations, funding options and financial management best practices. Engaging with online courses, workshops and financial mentors can provide valuable insights and help stay abreast of emerging trends. ... In today's digital age, technology is a major driver of business innovation and efficiency. Entrepreneurs must stay informed about the latest technological advancements relevant to their industry. This encompasses the implementation and utilization of new software, tools, and platforms to streamline operations, enhance productivity, and improve customer experiences.


Software Architecture in an AI World

Programming isn’t software architecture, a discipline that often doesn’t require writing a single line of code. Architecture deals with the human and organizational side of software development: talking to people about the problems they want solved and designing a solution to those problems. That doesn’t sound so hard, until you get into the details—which are often unspoken. Who uses the software and why? How does the proposed software integrate with the customer’s other applications? How does the software integrate with the organization’s business plans? How does it address the markets that the organization serves? Will it run on the customer’s infrastructure, or will it require new infrastructure? On-prem or in the cloud? How often will the new software need to be modified or extended? ... Every new generation of tooling lets us do more than we could before. If AI really delivers the ability to complete projects faster—and that’s still a big if—the one thing that doesn’t mean is that the amount of work will decrease. We’ll be able to take the time saved and do more with it: spend more time understanding the customers’ requirements, doing more simulations and experiments, and maybe even building more complex architectures.


Edge AI: Small Is the New Large

The technologies driving these advancements include AI-enabled chips, NPUs, embedded operating systems, the software stack and pre-trained models. Collectively, they form a SoC - system on chip. Software, hardware and applications are key to enabling an intelligent device at the edge. The embedded software stack in the chip brings it all together and makes it work. Silicon Valley-based embedUR specializes in creating software stacks for bespoke edge devices, acting as a "software integrator" that collaborates closely with chip manufacturers to build custom solutions. "We have the ability to build managed software, as well as build individual software stacks for small, medium and large devices. You can think of us as a virtual R&D team," Subramaniam said. ... OpenAI released a smaller version of the ChatGPT language model called GPT-4o mini, set to be 60% cheaper than GPT-3.5. But smaller does not mean less powerful, in terms of AI processing. Despite their smaller size, SMLs possess substantial reasoning and language understanding capabilities. For instance, Phi-2 has 2.7 billion parameters, Phi-3 has 7 billion, and Phi-3 mini has 3.8 billion.


Reflecting on Serverless: Current State, Community Thoughts, and Future Prospects

The great power of serverless is that starting with and becoming productive is much easier. Just think how long it would take a developer who has never seen either Lambda or Kubernetes to deploy a Hello World backend with public API on both. As you start building more realistic production applications, the complexity increases. You must take care of observability, security, cost optimization, failure handling, etc. With non-serverless, this responsibility usually falls on the operations team. With serverless, it usually falls on developers, where there is considerable confusion. ... Issues like serverless testing, serverless observability, learning to write a proper Lambda handler, dealing with tenant isolation, working with infrastructure as code tools (too many AWS options—SAM, CDK, Chalice, which one to choose and why?), and learning all the best practices overwhelm developers and managers alike. AWS has published articles on most topics, but there are many opinions, too many 'hello world' projects that get deprecated within six months, and not enough advanced use cases. 



Quote for the day:

"You are the only one who can use your ability. It is an awesome responsibility." -- Zig Ziglar

No comments:

Post a Comment