Pinterest Engineering’s Post

I enjoyed listening to this interview with Andrej Karpathy, one of the founding members of OpenAI, and whose youtube videos have helped me greatly in understanding the underlying architecture of large language models. One of the things I love is that "building on-ramps" for others is part of the mission and not incidental. As strides are made with new technologies it's important to help bring parity of use and knowledge of that technology. The more we do so, the more we bet on the numbers of humanity rather than the few. Diversity of thought helps to look at problems through different lenses, and sometimes to just be ignorant enough to ask a stupid question that others may not have considered or analogically translate a concept from one domain to a dissimilar domain. Fear of perception inhibits our ability to ask "dumb" questions and learn. LLM's don't have perception of your intelligence, and the removal of that barrier from gaining knowledge will create an entirely different world for my children to grow up in, and (optimistically) help bring parity of knowledge to the world. https://lnkd.in/gD4cUvne

Embrace a smarter learning journey on Hugging Face AI! Kickstart a new era of knowledge with Interview Kickstart's webinar led by Mina Ghashami, Applied Scientist at Amazon and Adjunct Lecturer at Stanford. Discover the power of Q&A models, dive into hands-on coding practice, and gain invaluable insights into top-notch question-answering practices. Don't miss out—register now and amplify your learning experience!  https://bit.ly/3tDHwNN, #aialgorithms #artificialintelliegence #artificialintelligenceai #machinelearning #machinelearningalgorithms #edutech #onlinelearning #onlinelearningplatform #huggingface #questionandanswer

Usually, I focus my #aiart time on anthropomorphizing cats because that normally yields the highest ROI for me (currency: funsies), but I'm impressed with this handling of an allegory. Microsoft Designer did a great job interpreting and visualizing my ideas about #ethicalAI and the implications of binary thinking during programming. So far, so good studying AI, and this was a fun challenge from Ronnie Sheer in the "Introduction to Prompt Engineering" course to generate a movie poster about the future of AI. #learntechwithlinkedin #promptengineering #ai #generativeart #creativeai

There is no such thing as "best practices" or a playbook when working with large language models. One of my favorite things about working in the AI space is that the tech we're dealing with every day didn't exist when I was first studying computer engineering. Every day we touch on uncharted territory. Which is why I couldn't be more excited to announce that the team at Outset is launching the world's fastest fully-interactive voice-powered AI user research interviewer! A huge congratulations to the engineering team (especially Nils Caspar) for putting so much thought, ingenuity, and hard work into making this product happen. As with all AI-moderated research, you can get rich understanding of your users at a speed and scale that was never possible before. But now, your participants can feel like they’re in an interview with a person with a human-realistic voice, the ability to interrupt the moderator when you’ve got more to say, and an interface designed for a free-flowing conversation. Check out the video below to see Aaron Cannon explain more and go to our product hunt launch to give us feedback!

Open-Source and Community-Focused: Mixtral is released under the Apache 2.0 license, making it an open-weight model. Its open-source nature encourages community involvement, fostering innovation and allowing developers to customize the model for specific needs. Pre-training and Fine-Tuning: The model is pre-trained on data extracted from the open web, with experts and routers trained simultaneously. Mixtral can also be fine-tuned for specific tasks, such as instruction following, where it achieves impressive performance. In summary, Mixtral represents a significant advancement in AI model development. It combines a novel architecture with high efficiency, multilingual support, and a strong focus on community and open-source principles. Llama 2 An AI model developed by Meta AI designed to enhance the capabilities and safety of language models. Here are the key features and aspects of Llama 2 based on the context documents: Updated Architecture and Increased Size: Llama 2 is an updated version of its predecessor, Llama 1, featuring improvements such as a 40% increase in the size of the pretraining corpus, doubled context length, and the adoption of grouped-query attention. The model is available in variants with 7B, 13B, and 70B parameters, although the 34B variant is mentioned but not released. Focus on Safety and Responsible Use: Llama 2 has undergone extensive safety testing and tuning, including the use of techniques like Supervised Fine-Tuning (SFT) and Reinforcement Learning from Human Feedback (RLHF) to improve safety alignment. The model has been evaluated for safety across various benchmarks, demonstrating improvements in truthfulness, toxicity reduction, and bias mitigation compared to its predecessor. Pretraining and Fine-Tuning Methodologies: The model was pretrained on a diverse mix of publicly available data, following Meta’s standard privacy and legal review processes. It was also fine-tuned for specific applications, such as dialogue use cases, to optimize its performance and safety for those scenarios. Open Release for Research and Commercial Use: Llama 2 and its fine-tuned variant, Llama 2-Chat, have been released to the public for both research and commercial applications. This open release aims to benefit society by enabling broader access to advanced language model technology while emphasizing the importance of responsible deployment. In summary, Llama 2 represents a significant advancement in language model technology, focusing on improved architecture, safety, and accessibility for a wide range of applications. Comparing the Models While Gemma, Mixtral, and Llama 2 all aim to propel AI forward, their distinct paths reflect the diverse priorities and challenges in the field:

Another one from our Same Day Skunkworks 💪 Test your assumptions about the meaning of ambiguous writing with Hidden Biases and see if you think like the crowd, AI or the people in power. Want to explore more toys built as MVPs in a hurry 🏃♂️ 🏃♀️ by our devs, designers and data scientists? See link in comments to explore. 👇 https://lnkd.in/gTr6d7P9 #AI #Gemini #chatgpt #mvp #development

Making your plans for #ILTACON24? The lineup on Thursday is full of great sessions – you're going to want to stay! Here's a session I'm coordinating that I can't wait to see: Customizing LLMs: Leveraging Technology to Tailor GenAI   Use the latest advances in LLMs to create #GenAI systems that leverage your organization's data. Dive into techniques like fine-tuning, Retrieval-Augmented Generation (RAG), and Small Language Models (SLMs). Learn which approaches to apply to your use cases to reduce hallucinations, optimize performance, and lower costs. We have an incredible lineup of resources for this in-depth session: Jeremy Pickens, Applied Research Scientist, Redgrave Data Danielle Benecke, Founder/Director, Baker McKenzie AI/ML Zonghui Wei, Senior Data Scientist, Baker McKenzie James Howard, Principal AI Engineer, Laurel Lauren Rothrock, Chief Product Officer, Litify Jaime Basilico, Principal Data & AI Technology Specialist, Microsoft #WeAreILTA

Interested in predictive AI, data and implementation with a bit of humour included? Tad Salyards has some great insights to share in his article "Lessons from the AI trenches" 😀 #aiimplementation #ai #predictiveai #implementation #data

Read my new blog here: https://buff.ly/3V9e4KD Imagine a world where technology does more than just follow commands—it anticipates your needs and simplifies your life. This isn't science fiction; it's the reality we're stepping into with the advent of Large Language Models (LLMs). In this blog, I'll share how these advanced AI systems are not only revolutionizing my field of software engineering but also how they're set to transform the tech in your everyday life. Whether you're a tech enthusiast or just curious about the future, join me in exploring the profound impact LLMs are having on the way we work, play, and interact with the world around us. #techlife #techblogger #futurology