Google Maps Platform’s Post

This is really cool. Having LLM's debate each other to improve result quality. Turns out that every LLM doesn't have to talk to every other one -- sparse topologies work great. I wonder what would happen if you weakened the individual LLM's but greatly increased their numbers, and allowed them to communicate with a "Small World Network" topology https://lnkd.in/dwJEd33W which are sparse topologies where every node is log(N) steps away from every other one (very common in the brain). Think "6 degrees of separation". (you would probably need to retrain the individual LLM's so they were each specialized in individual topics to fit within their smaller capacity)

A very interesting paper! Paper: https://lnkd.in/gfA_JtGV Abstract: Multi-agent debate has proven effective in improving large language models quality for reasoning and factuality tasks. While various role-playing strategies in multi-agent debates have been explored, in terms of the communication among agents, existing approaches adopt a brute force algorithm -- each agent can communicate with all other agents. In this paper, we systematically investigate the effect of communication connectivity in multi-agent systems. Our experiments on GPT and Mistral models reveal that multi-agent debates leveraging sparse communication topology can achieve comparable or superior performance while significantly reducing computational costs. Furthermore, we extend the multi-agent debate framework to multimodal reasoning and alignment labeling tasks, showcasing its broad applicability and effectiveness. Our findings underscore the importance of communication connectivity on enhancing the efficiency and effectiveness of the "society of minds" approach. #aiagent #ml #LLMs #paper

Had the opportunity to attend Google ISV event in London,(presenters from Nvidia, ElasticSearch, Glean, Anthropic, Weight &Bias) couple if interesting things from that, 1. Glean knowledge graph of the company which if it really works, it helps teams to look for the information in one place (https://lnkd.in/eEYAABYC) 2. Anthropic map the minds (https://lnkd.in/e4wQ39wK) How the LLMs internals are being explored and make the LLMs safer.

🚀 New Blog Post Alert! 🚀 I'm excited to announce my latest blogpost, written in collaboration with neptune.ai! In this post, I explore the challenges and strategies for deploying Large Language Models locally. I discuss key topics like cost savings, privacy benefits, and memory optimization using techniques such as quantization and flash attention. I also highlight tools and libraries like Llama.cpp, Ollama, LM Studio, Unsloth AI, and WebLLM, along with best practices for using orchestration frameworks or routers like LangChain and TensorOps' LLMStudio. If you're considering running LLMs on your own hardware, this is worth a read. Let me know your thoughts! 👉 https://lnkd.in/dfRA5fFN

I'm just goofing off at this point.. Happy Thanksgiving y'all! For more serious conversation about building AI-native features into your application, check out the full episode this clip is from 🙃 https://lnkd.in/g_7gQAGa

🚀 61% of engineering teams are already on board with Generative AI! Curious why and how to join them? 🤔 Check out our latest blog to see how AI is transforming software engineering and what it means for your team. Link in comments.

🎥 Raid AI: Tackling Deepfake Audio 🎙️ In this video, we showcase the problem, present our innovative solution, and give you a sneak peek of our website in action—featuring a deepfake of our “close friends” Obama, Trump, and Morgan Freeman. Discover how we’re leading the way in deepfake detection and offering deployment options tailored to your infrastructure. Curious to see it in action? Watch the demo now! 🔗

Sometimes evals are all you need. That's why Arize AI and Ragas have setup a recurring meetup dedicated to exploring and advancing OSS tools and best practices around it. https://lu.ma/llm-evals Come hang out with some of the best in the industry: Shahul ES Jithin James and Jason Lopatecki!

Exciting developments in the world of artificial intelligence are shaping the future of software engineering. Our latest blog post dives into a thought-provoking discussion from the Hacker News community, where software engineers are seeking recommendations to enhance their understanding of AI and its applications. By exploring various resources and insights shared in this thread, you can elevate your knowledge and stay ahead in this rapidly evolving field. Join the conversation and discover valuable recommendations that could help you on your journey. Read more here: [Hacker News Discussion on AI Recommendations](https://ift.tt/xtGFDwh).