--- title: "AI stories scarier than realizing you clicked 'Reply All'" newsletter: "MLOps Community" date: 2024-07-18 source: https://aaif.live/newsletters/mlopscommunity/2024-07-18-ai-stories-scarier-than-realizing-you-clicked-reply-all --- # AI stories scarier than realizing you clicked 'Reply All' *Plus, evaluating beyond benchmarks, creating characters, cost saving deployment, and hidden gems.* *MLOps Community — Agentic AI Foundation, 2024-07-18* ## Extending AI: From Industry to Innovation // Sophia Rowland & David Weik // MLOps Podcast #246 Now Independence Day is over, with a mere three months to go, expect the stores to be full of Hallowe’en stuff. Not one to be left behind, we’ve got a podcast where Sophia and David share some of the horror stories they’ve encountered, like zombie models that are still running but not being used. We also chat about real-time integration of models in fraud detection and generative AI deployment, and their work on improving safety and efficiency in industrial settings using edge computing and computer vision models. Don’t miss this episode, it’s a scream! Video [https://home.mlops.community/home/videos/extending-ai-from-industry-to-innovation] || Spotify [https://open.spotify.com/episode/4XFrj6EFUqoDkundJldTDa?si=7hBB8vVXRkCdTAMA7jf-BQ] || Apple [https://podcasts.apple.com/us/podcast/extending-ai-from-industry-to-innovation-sophia-rowland/id1505372978?i=1000662036964] [https://podcasts.apple.com/us/podcast/extending-ai-from-industry-to-innovation-sophia-rowland/id1505372978?i=1000662036964](https://podcasts.apple.com/us/podcast/extending-ai-from-industry-to-innovation-sophia-rowland/id1505372978?i=1000662036964) [https://home.mlops.community/home/videos/extending-ai-from-industry-to-innovation](https://home.mlops.community/home/videos/extending-ai-from-industry-to-innovation) [https://open.spotify.com/episode/4XFrj6EFUqoDkundJldTDa?si=7hBB8vVXRkCdTAMA7jf-BQ](https://open.spotify.com/episode/4XFrj6EFUqoDkundJldTDa?si=7hBB8vVXRkCdTAMA7jf-BQ) ## VIRTUAL TALK FROM TECTON Data is crucial for AI/ML systems but often becomes a bottleneck in development. Data scientists and engineers grapple with the complexity of building and maintaining feature pipelines, ensuring consistency, data freshness, and achieving real-time performance. In this talk, Sergio Ferragut, Principal Developer Advocate at Tecton, will explore how declarative frameworks are a game changer for production AI, empowering data scientists and ML engineers to collaborate effortlessly and build production features with ease. Sergio will demonstrate how Tecton’s declarative framework streamlines collaboration between data scientists and ML engineers, automates feature materialization, and handles diverse data types. Discover strategies to promote feature reusability, eliminate training-serving skew, and simplify complex feature creation. He’ll also discuss how these frameworks drive automation of production-ready pipelines, accelerating AI projects and making AI-powered applications smarter. Key topics include: * Overcoming traditional feature engineering challenges * Real-world applications and success stories * Strategies for efficient feature reusability and scalability Register Here [https://resources.tecton.ai/supercharge-production-ai-with-features-as-code] ## Evaluating the Effectiveness of Large Language Models // Aniket Singh // MLOps Podcast #248 For Aniket, MLOps is more than just a job; it’s Meaningful Learning Opportunities. He talked me through the research he does outside of his day job, including evaluating LLMs beyond traditional benchmarking and conducting stealth assessments using confidence scores. He stressed the importance of evaluating LLM capabilities in real-world applications, mentioning experiments on multi-agent workflows and the impact of feedback on model confidence. His goal is to improve LLM reliability and efficiency in production through multi-agent and multi-modal approaches. Be sure to listen for LLM - Lifelong Learning and Mastery. Video [https://home.mlops.community/home/videos/evaluating-the-effectiveness-of-large-language-models] || Spotify [https://open.spotify.com/episode/3zWOw5o1hqooWi7Hzvl8ei?si=ERnm0X3DQFimXDy4tBaKaQ] || Apple [https://podcasts.apple.com/us/podcast/evaluating-the-effectiveness-of-large-language/id1505372978?i=1000662431770] [https://podcasts.apple.com/us/podcast/evaluating-the-effectiveness-of-large-language/id1505372978?i=1000662431770](https://podcasts.apple.com/us/podcast/evaluating-the-effectiveness-of-large-language/id1505372978?i=1000662431770) [https://home.mlops.community/home/videos/evaluating-the-effectiveness-of-large-language-models](https://home.mlops.community/home/videos/evaluating-the-effectiveness-of-large-language-models) [https://open.spotify.com/episode/3zWOw5o1hqooWi7Hzvl8ei?si=ERnm0X3DQFimXDy4tBaKaQ](https://open.spotify.com/episode/3zWOw5o1hqooWi7Hzvl8ei?si=ERnm0X3DQFimXDy4tBaKaQ) ## Job of the Week [https://jobs.lever.co/tecton/6088fcae-ae95-4944-a412-26c3a1cac2b6](https://jobs.lever.co/tecton/6088fcae-ae95-4944-a412-26c3a1cac2b6) ## Guiding NPC Agent Behavior // Nyla Worker // IRL Meetups San Francisco #86 ## MLOps Community IRL Meetup From the person who drinks yerba mate from a traditional gourd during meetings to the one who uses a standing desk with a treadmill and codes in Julia while walking, every office has its characters. If you work remotely and miss that experience, this chat covers creating virtual characters using RAG knowledge bases, embodied in 3D environments via Unreal or Unity. It details workflows integrating speech-to-text, fine-tuned LLMs, text-to-action/animation, and text-to-speech. Use cases span gaming and education, emphasizing immersive, accurate interactions. It highlights input moderation, model fine-tuning, and AI audits to maintain integrity. Customization tools for character backstories, personalities, and narrative nodes ensure scalability and detailed management, with continuous testing for validation. PSA: If you think your office doesn’t have any ‘characters,’ it could be you. Watch it here [https://home.mlops.community/home/videos/guiding-npc-agent-behavior] [Watch it here](https://home.mlops.community/home/videos/guiding-npc-agent-behavior) AWS Inferentia and AWS Trainium deliver lowest cost to deploy Llama 3 models in Amazon SageMaker JumpStart [https://home.mlops.community/home/blogs/aws-inferentia-and-aws-trainium-deliver-lowest-cost-to-deploy-llama-3-models-in-amazon-sagemaker-jumpstart] They say a penny saved is a penny earned. Well, with this blog you’ll be cashing in as it explains how AWS integrates Llama 3 models with Trainium and Inferentia to cut deployment costs by up to 50%. It guides you through deployment via SageMaker Studio and Python SDK, emphasizing no-code and pre-configured options for ease. Detailed instructions and code examples are provided for building chatbots and AI assistants, highlighting the benefits of AWS's high-performance accelerators for efficient management and scalability. Just remember who helped you save when you’re rolling in the dough. With thanks to AWS for their contribution. [AWS Inferentia and AWS Trainium deliver lowest cost to deploy Llama 3 models in Amazon SageMaker JumpStart](https://home.mlops.community/home/blogs/aws-inferentia-and-aws-trainium-deliver-lowest-cost-to-deploy-llama-3-models-in-amazon-sagemaker-jumpstart) ## Hidden Gems ## IRL Meetups Lisbon [https://lu.ma/7sowmead] - July 18 New York [https://lu.ma/6v4jcdnt] -July 18 Denver [https://mlops.community/event/pizza-pints-expert-speaker-intro-to-machine-learning-toolkits-with-kubeflow/] - July 30 Bucharest [https://lu.ma/gbi3flph] - July 31 [Lisbon](https://lu.ma/7sowmead) Thanks for reading. See you in Slack [https://go.mlops.community/slack], YouTube [https://www.youtube.com/channel/UCG6qpjVnBTTT8wLGBygANOQ?view_as=subscriber], and podcast [https://home.mlops.community/public/content/] land. Oh yeah, and we are also on X [https://twitter.com/mlopscommunity]. The MLOps Community newsletter is edited by Jessica Rudd [https://www.linkedin.com/in/jmrudd/]. --- Source: https://aaif.live/newsletters/mlopscommunity/2024-07-18-ai-stories-scarier-than-realizing-you-clicked-reply-all