Yonas Yohannes’ Post

Interesting interview with the CIA on Generative AI, lots of great Q&A. One of my favorites is "Can you name your large-language model partners?" Answer: As a smart customer, we are not tying our boat to a specific set of LLMs or a specific set of vendors. We are evaluating and using practically all the high-runner LLMs out there, both commercial-grade and open source. We are not viewing the LLM market as a singular one where a single lab is better than the others. https://lnkd.in/eBkUKBJW

Have you heard about RAG? Retrieval-Augmented Generation (RAG) integrates real-time, verified data into generative AI language models. Why RAG matters? ▶️ It fetches real-time data so the responses are correct and relevant. ▶️ It reduces error count, checking data against real facts from a knowledge base. ▶️ It allows to update LLM's knowledge continuously without needing frequent retraining. In tomorrow's Smart Moves chapter, we'll discuss the idea behind it and give you some ideas for using RAG. Stay tuned!

How to Overcome Extracting Answers from Large Document Sets with LLMs read more here https://lnkd.in/ge4eE5FN 🌟 How to Overcome Extracting Answers from Large Document Sets with LLMs 🌟 Navigating through enormous document sets to extract precise answers can be a monumental task. With Atlas AI at CloudScale, we're simplifying this challenge using Large Language Models

OpenAI has just released two groundbreaking models, o1 and o1-mini, both outperforming all others by a significant margin. I believe this is one of those pivotal moments where we need to reevaluate our perspective on Generative AI. • AI can now perform more complex tasks than ever before, including many that previously required human intervention. • All development, especially in AI, will accelerate. We’ll witness an even faster pace of change moving forward. • Our interaction with AI is evolving, and the gap between human+AI and human-only work has widened. The difference in intelligence and creativity is becoming so vast that human-only work isn’t really an option anymore. It’s almost a bit daunting, but I’m thrilled to think about everything that’s now possible—things that just weren’t feasible two days ago. 😊 (In the graph, note that gpt4o was the model that was previously leading) (it's actually preview versions, o1-preview and o1-mini-preview, I left it out of this post just to make it easier to consume. The real o1 model will be even better).

Industry expert, Venkatesh Padmanabhachari says that as long as human judgment is needed for 30% or more of the tasks one handles, machines cannot replace humans, but they can definitely augment, advance, and accelerate their work. Read on to know more...

Big thanks to Mike Lawson of CUBroadcast for having me on the show again to talk about how the financial sector is adapting to generative AI! Mike is a great interviewer and always has the pulse of financial services trends. Give it a listen for insights on how credit unions can establish a foundational strategy and adaptive risk framework to enable safe adoption and value realization. https://lnkd.in/gwFTn6eW

Join AI/ML Architect Richard Chang from Salesforce for Part 2 of our series on the latest trends in inference optimization for Large Language Models (LLMs)! Building on our previous discussion of serving layers and speculative decoding, we'll dive into quantization and multi-adapter model deployment strategies. Learn how to solve real-world problems, explore these cutting-edge technologies, and get actionable insights on leveraging Amazon SageMaker's Model Optimization job and multi-adapter LoRA deployment mechanisms to stay ahead in generative AI!

Provides key insights into the fundamentals of how best to utilize generative AI for FP&A activities.

One of our most sought after services is designing AI that works for you in real time. The old pattern of crunching data once a day, or any other arbitrary interval, seems simpler than real-time technology-wise, but is actually an unnecessary bottleneck in many cases. Just a year ago data for AI applications was mostly demanding to collect. With large language models, business critical data is suddenly everywhere, published all the time everywhere around you. Why let that potential go to waste? The below blog post discusses real time AI technologies at length, have a read :)

AI projects generally fall into two categories: 1) Simple ones that require an API, a pretrained model, or direct generative AI. 2) Complex ones that demand a deep understanding of data science, problem-solving skills, and a lot of mind, where the solution is like a masterpiece art.