Avijit Gupta

Many of these techniques were completely new to me, and I foresee them being very useful: https://red.ht/4bBblya #ML #DL #NN #oss #opensource

4

AlloyDB provides 2x better price-performance than self-managed PostgreSQL. See how in this article from Sridhar Ranganathan and Shesh Ranganath.

14

#Rise of Micro LLMs: A Roadmap to Business Success with Generative AI In the current AI landscape, Large Language Models (LLMs) like GPT-3, BERT, and T5 have garnered significant attention for their impressive performance on various NLP tasks, including text generation, question answering, summarization, and translation. However, the training and deployment of these massive, general-purpose models come at a substantial cost, often reaching up to $10 million. These expenses cover not only computational resources but also specialized hardware such as GPUs and the necessary software infrastructure. Micro LLMs are developed by refining larger models to address industry-specific, domain-specific, and customer-specific scenarios, offering several key advantages: #Personalization: Enhances user engagement and satisfaction by providing more relevant interactions, leading to higher retention rates. #Efficiency: Requires less computational power and memory, making them suitable for deployment on smaller devices or in edge computing scenarios. #Accuracy: Reduces hallucinations and bias by focusing on enterprise-relevant data. #Cost-effectiveness: Eliminates the need for expensive data scientists or costly hardware and systems. While general-purpose or "macro" LLMs are beneficial in scenarios requiring a broad understanding, their wider scope may not always align with the specific nuances of enterprise-focused tasks. Additionally, their flexibility and adaptability can sometimes come at the expense of precision compared to Micro LLMs. #Benefits and #Limitations Micro LLMs offer significant benefits, such as being trainable on modest datasets and having simplified architectures that enhance interpretability. Their compact size makes them ideal for deployment on mobile devices, and their ability to process data locally is particularly valuable for IoT edge devices and organizations with stringent privacy and security needs. However, their limitations include more restricted knowledge bases due to smaller training datasets and a less comprehensive understanding of language and context, which can lead to less accurate and nuanced responses compared to larger models. #The Future of #Micro LLMs The future of small language models looks promising with increased accessibility and cost efficiency, making them attractive for businesses and individuals. Their lower computational requirements enable real-time processing on edge devices, enhancing privacy and energy efficiency. These models are easier to fine-tune for specific tasks, ensuring effective and specialized solutions. Additionally, their faster response times and seamless integration with existing systems improve user experience. As advancements continue, small language models will become more capable, promoting the democratization of AI and fostering innovation across various fields. #GenerativeAI #Micro-LLMs

20

✨ Can regular expressions determine if a number is prime? 😲 For all the math enthusiasts and regex aficionados, here’s an eye-opener: a unique, pattern-based approach to prime checking! 🧩💡 I stumbled across this fascinating read that reimagines prime checking through regular expressions—an angle you don’t see every day. Check out this article to explore how unconventional tools can redefine familiar problems. 🔗 Take a moment to dive in and see what creative solutions can emerge from unexpected places. Shared with you by Skill Alley! https://lnkd.in/eiFWehxe #SoftwareEngineering #CodingTips #RegularExpressions #PrimeNumbers #MathInTech #SkillAlley #TechInsights #Developers #InnovationInCode #skillalley

18

LLMs and Optimization: An Unbeatable Combo!

19

Very cool article by Christoph Molnar on the notion of inductive bias (here for Random Forests). The concept of inductive bias is often overlooked by practitioners, who assume this is a purely theoretical notion. On the contrary! This is one of the most fundamental aspects to consider when picking a model, interpreting its predictions or troubleshooting an underperforming system. #inductivebias #cart #trees #machinelearning #troubleshooting

1

Check out the latest in our "Debuggability" blog series! This blog is authored by Kevin Downey Jason Johl and me and talks about how we built a Debuggability Paved Road at Intuit using Ephemeral Containers and Argo Workflows. We'd love to hear your thoughts on the topic. Also, what tools do you build as-a-platform to help debug service, compute and network issues in Kubernetes at scale? #intuit #kubernetes #debug

22

To lift my reply up, I have expanded it here. These topics form much of the material on Day 2 of my Practical Messaging workshop next NDC Oslo 10-11 June) So the first thing to think about is the type of asynchronous conversation: * Messaging (uses a queue; transfer of control) * Eventing (uses a stream; publish-subscribe) (Replies can be either stream or queue, default to queue unless you are broadcasting). We also need to think about Provider over Producer/Consumer when thinking about the interaction. Who provides the API? The Consumer provides the API for messaging; the Producer provides the API for eventing. Concepts like In-Out, Out-Only are useful here from the old WS- Messaging specs. I use that taxonomy when I teach my workshop because it is still a useful tool. For example in In-Out we have a consumer that acts as the API provider and receives a message; it should use a queue; we have a response that probably travels back over that queue. The main issue that brings this up is causality - I need this step (taking a payment) to occur before this step (place an order). Definitely seen a few systems struggle with “event chaining” where they lose causality, particularly pernicious when a component fails (and can lead to nasty loops where work goes round and around - imagine folks keep turning up with the same order to your house). I spend time in my trying to avoid everyone reaching for a Process Manager (Saga) here. In a lot of cases I recommend what I call Turn-Taking which is essentially Pat Helland’s Activity model (components co-operate via their internal business logic).

32

4 Comments

DynamoDB Demystified: Mastering NoSQL Database Design: Introduction to NoSQL and DynamoDB: The evolution of databases brought us from the structured realms of SQL to the dynamic landscapes of NoSQL to cater to more complex, scalable applications. NoSQL databases prioritize how data is retrieved rather than how it’s stored, with DynamoDB and MongoDB leading the charge. One fundamental principle guiding NoSQL adoption is the CAP Theorem, wherein DynamoDB excels in Availability and Partition tolerance (AP), but not in Consistency. Setting Up Tables and Querying in DynamoDB: Creating a DynamoDB table primarily involves setting up a primary key, which includes a hash key and an optional sort key. The schema is flexible, reflecting the database’s loose structure that accommodates a variety of data types without predefined constraints. Queries in DynamoDB can be performed using the hash key alone or in combination with the sort key. Additionally, filter expressions can refine results, though costs are incurred based on the entire query volume, not just the filtered results. Indexes and Design Considerations: Understanding when to use Global Secondary Indexes (GSIs) and Local Secondary Indexes (LSIs) is crucial. These indexes facilitate efficient searches beyond simple primary key queries but come with additional cost implications as they duplicate data to speed up access. When designing a DynamoDB solution, consider the primary search requirements: if your application frequently retrieves large datasets, DynamoDB might not be optimal due to its design and cost model. Operational Insights and Best Practices: DynamoDB operates under an eventual consistency model, meaning immediate consistency of data across all nodes is not guaranteed after a write operation. If your application requires up-to-the-second data accuracy, you’ll need to use stronger consistency reads, which double the cost. Moreover, the architecture discourages traditional relational database designs, pushing developers to think in terms of large, denormalized tables rather than separate interconnected ones. For managing data, be aware of the item size and result set limits — each item can be up to 400Kb, and query result sets are capped at 1Mb, which can affect performance and design strategies. Conclusion: DynamoDB’s Fit in Modern Applications: While DynamoDB offers powerful capabilities for managing NoSQL data, it’s not suitable for every scenario. It excels in environments where high write and read speeds are crucial and where data relationships are not overly complex. For large-scale data extraction and reporting, consider integrating DynamoDB with other database solutions to leverage its strengths without becoming constrained by its limitations. Understanding these aspects will help you make informed decisions about when and how to use DynamoDB effectively in your projects. #softwareengineering #NoSQL #DynamoDB #AWS #programming

1

Kicking off a super interesting series on the internals of DragonflyDB - a multi-threaded drop-in replacement of Redis ⚡ On a single machine, it offers 25 times the throughput of Redis as it is optimized to leverage every single ounce of performance from the underlying hardware. The first video of the series is out and it is a quick architecture overview and covers two key design decisions. give it a watch - youtu.be/XbV1LoVsbME The low-level design decisions and internal details will just blow your mind away - right from their design of DashTable (not hash table) to sorted set implementation. Trust me, your perspective of looking at Data structures will change after the series :) ⚡ Subscribe to the channel and follow along; I keep things no-fluff. youtube.com/c/ArpitBhayani #AsliEngineering #SystemDesign #DatabaseInternals

286

5 Comments

Our trajectory with Google Cloud epitomizes a transformative strategic pivot. What began as a challenging market positioning has evolved into a compelling narrative of operational resilience and technological leadership. Over seven pivotal years, we've fundamentally reimagined our enterprise value proposition—transitioning from a peripheral cloud solution to a mission-critical AI innovation partner. The financial metrics tell a powerful story: we've systematically converted recurring quarterly deficits into a robust framework for sustainable profitability. Buckle up the next 7 years are going to be epic @ Google Cloud! 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 🚀 #google #googlecloud #GCP #gcpcloud

20

When APIs for older LLMs are no longer supported, you’ll need to switch to newer APIs. If you're using LLMs in your application to get structured responses like JSON etc, switching models with the same prompts may not be as straightforward as it seems. Many developers assume that moving from one model to another (e.g., from GPT-3.5 to GPT-4) will be easy, but it can throw some curveballs—such as unexpected outputs and often, a significant price hike. Pro tip: Always have a backup plan for your LLMs. Consider this approach: Keep a more budget-friendly LLM alternative as a backup. It can be from other LLM providers or even open-source models. Whatever backup you choose should be thoroughly tested. This might seem like extra work now, but it can save you when things go wrong. Have you tried this? What worked for you? #llm #AI #gpt

42

3 Comments

🤔 Interesting perspective from Infosys Chairman Nandan Nilekani: "Let Silicon Valley build LLMs; India should lead in AI use cases." At Meta's Build with AI summit, Nilekani suggested that India should focus on practical AI applications rather than competing in the LLM space. But is this the right approach? Key points to consider: • Silicon Valley giants are heavily invested in LLM development • India could focus on: - Synthetic data generation - Small language models - Real-world applications - Building robust data infrastructure The vision is for India to become the global "use case capital" for AI, emphasizing frugal innovation and practical solutions. What do you think? 💭 Should India stick to developing practical AI applications? 💭 Or should we also invest in fundamental AI research and LLM development? 💭 Could focusing solely on use cases limit our AI capabilities in the long run? Let's discuss! Share your thoughts in the comments below. 👇 #AI #DigitalIndia #LLM

30