Subodh Kumar

Excellent report from General Dynamics Information Technology John P. Curran regarding adoption and trends of AI in the Govt. I used Meta to generate summary below and included link to original article. Main AI use cases are chat and creating/updating computer code. Note also the 23% failure rate of AI projects due to user expectation misalignment. This can be overcome with thorough protoyping and establishing use case metrics. - Successful AI project completion timelines are narrowing, with an average of 14 months from pilot to production. - Future demand for AI projects is building around chat and coding-related functions. - 54% of surveyed AI experts said projects currently under development involve improving computer code. - 59% of projects involve "research and understanding", which may relate to creating chat functions. - 23% of AI projects fail due to misalignment between AI solutions and user expectations. - Partnership efforts are required for successful AI projects. Data diversity is crucial, and incorporating diverse data sources can lead to better AI outcomes. Trends: - AI adoption is increasing across the Federal government, with a growing focus on chat and coding-related functions. - Agencies are leveraging cloud-native services, tooling, and other capabilities to accelerate AI project development. - There is a growing interest in translating older code to newer languages, such as Python or cloud-based analytic tools. - Domain-specific chat functionality is becoming a priority, with many agencies planning to implement chat services in the next two years. Statistics: - 325 AI experts surveyed across the Federal government - 14 months: average time from pilot to production for AI projects - 54%: percentage of projects currently under development that involve improving computer code - 59%: percentage of projects that involve "research and understanding" - 23%: percentage of AI projects that fail due to misalignment between AI solutions and user expectations. Bill Casola John Grim Leslie Anne McIndoe Kim Giffin Michael S. Freeman III Tobias Quaranta https://lnkd.in/eqvmAVFG

3

Here is my column on the #TiEcon2024 conference in #indicanews:

61

4 Comments

Learn how to get up and running with Claude 3.5, the newest model from Anthropic on #VertexAI with Weights & Biases Weave! This tutorial is designed to help you get started with the Claude 3.5 Sonnet model on Google Cloud’s Vertex AI using Python. #aiml #generativeai #googlecloudpartners

3

AI Ramblings AI Musings Have you ever wondered about the reliability of large language models? Cleanlab, a dynamic AI startup originating from MIT's quantum computing lab, introduces an innovative solution - the Trustworthy Language Model. LLM's are renowned for their creative abilities but have often raised concerns about accuracy. With the introduction of the "Trustworthy Language Model", users can gauge the reliability of generated outputs on a scale from 0 to 1. The tool should empower users and businesses to make informed decisions, distinguishing between trustworthy responses and those that may not be accurate. In essence, it's like having a BS-o-meter for your chatbot! "Addendum: it's worth noting that the example in the story is from GPT3.5 and that GPT4 is much better, although still nowhere near perfect!" #AI #TrustworthyAI #Innovation #artificialintelligence #aimusings #airamblings #cleanlab

1

2 Comments

Language models are central to most AI applications and tools, but they can be huge and require large amounts of complex and expensive compute power to run. Most of the well known tools such as ChatGPT, Gemini and CoPilot use large language models that contain many trillions of parameters. In the case of GPT 4 that powers the latest model of ChatGPT it is rumoured to have around 100 Trillion. What is a parameter and why is it important? Parameters are the way in which the large language models interprets and is able to create sentences and combine words to solve problems. The parameters are what the model uses to determine the best output based on the input. The more parameters the more accurate the output can be. However, this comes with a trade off in that the more parameters the more compute power is needed to run the model. This is where small language models, or SLMs, can help. Small language models still often contain billions of parameters but are of a much smaller size that allows them to run on less powerful devices. For example you can take Microsoft’s latest Phi model and run it locally on a MacBook Pro. Why is this useful? Let’s say you want to run a model on a small edge device to interpret data to predict maintenance schedules from a bunch of sensors on a piece of machinery. The device is running in a remote location that does not have an internet connection are therefore prevents you from connecting to one of the LLMs. By downloading and training the LLM on a customised dataset you can have this running independently and with little hardware costs. Another use case is an internal chat bot where you do not want to connect your data to a public cloud platform but instead want to run it on your own hardware. Over the next few weeks I am going to be testing and using some of the small language models to try and solve a few problems and will be posting my progress here. #ai #llm #slm #languagemodels #artificialintelligence

7

4 Comments

One more conference talk at #DEML2024 on "Integrating GenAI with Current Data Infrastructure: LLM Fine-Tuning and Ecosystem" https://lnkd.in/ebr8Tw_m Thanks to all who attended. Once you register there should be a recording for it. Thanks, Mert Bozkir, for managing the session. #DataEngineering #MachineLearning #GenerativeAI #LLM #FineTuning #genAI

18

1 Comment

Learn more about the role of vector datastores in GenAI applications from Amazon Web Services (AWS)'s Rahul Pathak Ganapathy (G2) Krishnamoorthy & Vlad Vlasceanu in this very insightful blog! "Embeddings should be stored close to your source data. As a result, where you store your data today, as well as familiarity with these database technologies, scale in terms of vector dimensions, number of embeddings, and performance needs will determine which [vector database] option is right for you" --> This blog covers many great topics, including use of Amazon OpenSearch, Amazon Aurora PostgreSQL (with pgvector extension, Amazon RDS PostgreSQL (with pgvector extension), as well as details on using vector datastores for RAG and common considerations. Check it out! #AWS #database #vectordatabase #genai #RAG https://lnkd.in/eZ-_MN_Q

12

Innovating Manuscript Publication: A GenAI-Powered Platform Using Agentic RAG Architecture. For more insights on how Generative AI is transforming publishing platforms explore this article #GenAI #AI #GGU #ManuscriptPublication #AgenticRAG #RAG

14

2 Comments

This article looks at a strategic blueprint for CEOs and CIOs. #ceo #cio #cto #ai #techlandscape #strategy #options #template #blueprint https://lnkd.in/e-sYbPUK

3

I really enjoyed this conversation as it allowed us to explore the fascinating world of Small Language Models and uncover why they're becoming the true workhorses of enterprise GenAI tasks! 💡 Given how frequently these questions come up in my discussions, I thought I'd share some key insights: 1️⃣ Why are SLMs important for enterprises? 🏢 Enterprises are leveraging LLMs and other GenAI solutions to improve their business processes, aiming to boost efficiency without inflating costs. While larger LLMs offer impressive capabilities, they come with significant cost implications - especially the proprietary models. Enter Small Language Models: though they might not match their giant cousins in raw performance initially, they can catch-up (and often outperform) through customization. Here's the game-changer: a task-customized SLM can deliver similar or superior performance compared to proprietary LLMs while being orders of magnitude more cost-effective! 💪 2️⃣ What is LLM customization? 🎯 To customize your LLM, you'll need two essential tools: - A solution to transform your enterprise data (knowledge) into training data and/or generate task-specific (skills) training data 📊 - An efficient method to train your SLM on this curated data 🔄 3️⃣ What makes it hard? 🤔 Data, of course! While numerous solutions exist to "bake the data into the model," we lack enterprise-grade approaches for converting enterprise data into training data or generating skills data. Sure, you could turn to data vendors, but that route is typically extremely expensive and often struggles to scale as your enterprise data grows. 📈 4️⃣ What tools are available for LLM customization today? ⚡ This is exactly where InstructLab shines! ✨ Unlike other customization offerings, IL provides both tools in one comprehensive open-source package, backed by the expertise of RedHat and IBM. IL features some of the most advanced synthetic data generation methods, enabling you to: - Convert your enterprise data into high-quality training data - Synthesize skills data using open-source LLMs - Leverage novel large-scale training methods to optimize your small model The results? 🎯 On numerous enterprise tasks, including RAG implementations, IL-tuned 7B/8B models consistently outperform proprietary models at a fraction of the cost! 🤝 Let me know if this was helpful, I'll be happy to create a comprehensive long(-er) form post sharing concrete results from real-world implementations. #GenAI #MachineLearning #EnterpriseAI #Innovation #TechTrends #InstructLab #SmallLanguageModels

54

David Linthicum, former #Chief_Cloud_Strategy_Officer at #Deloitte_Consulting and founder of Linthicum Research, told Fierce that in many cases "#on-premises resources are going to be #more_cost-efficient, and considering that AI systems cost 2 to 3 times the cost to build and run, enterprises are not automatically sold on public cloud providers for AI." "#On-premises_systems are generally #cheaper, can be #more_secure, and feel like they provide #more_control; enterprises are opting for them more than the cloud providers would like," he added. "I suspect that the cloud providers will see good growth around AI, but enterprises will be repatriating the platform options, considering the #high_cost_of_the_cloud."

1

David Linthicum, former #Chief_Cloud_Strategy_Officer at #Deloitte_Consulting and founder of Linthicum Research, told Fierce that in many cases "#on-premises resources are going to be #more_cost-efficient, and considering that AI systems cost 2 to 3 times the cost to build and run, enterprises are not automatically sold on public cloud providers for AI." "#On-premises_systems are generally #cheaper, can be #more_secure, and feel like they provide #more_control; enterprises are opting for them more than the cloud providers would like," he added. "I suspect that the cloud providers will see good growth around AI, but enterprises will be repatriating the platform options, considering the #high_cost_of_the_cloud."

2

Learn how DDN helps NVIDIA Cloud Providers power AI for their customers to train LLMs and build the next generation of GenAI apps https://lnkd.in/gSCqST9e

36

🌍 A groundbreaking #AI model fuses atmospheric & oceanic data for unmatched #weather & #climate prediction accuracy, enabling precise long-term forecasts for disaster prep, resource planning, & resilience. 🌦️ 🎥 Learn more ➡️ https://nvda.ws/4ft64LV

3

Can we have LLM routers? Unify has created a solution which abstracts the LLM being used - allowing us to use the best of breed. https://unify.ai/ An interesting concept to combine multiple LLMs #llms #genai #generatieveai #capgemini #llm #opensourceai #capgeminiindia #ai #artificialintelligence #software

6

1 Comment