Consig AI

Some thoughts on the impact and rise of non-determinism in enterprise computing. This shift is already transforming how we approach AI and automation, bringing both opportunities and challenges. Agree? Disagree? I’d love to hear your thoughts.

TADSummit Podcast 76: Truth in AI, Consig AI, RJ Burnham https://lnkd.in/eaQua3Y5 RJ has been working on #Voice #AI for over 25 years! #AI being the latest tool to enhance voice communications. RJ will be presenting at TADSummit 2024, October 22-23 on "Voice AI: Breaking Passed the Bullshit" We discuss his learning in voice AI, all the way back to his Voxeo days, the importance of #trust and how AI #hallucinations are impacting trust today. And ways to solve that issue as #RAG (Retrieval-Augmented Generation) limited responses are not as good as when the #LLM (Large Language Model) is allowed free reign. We discuss fuzzy rules and the cost of breaking rules given context. Strolid, Inc., Voxist, Thomas McCarthy-Howe, Karel Bourgois, Noah Rafalko, TSG Global, Inc., Vin Micciche, Le Voice Lab, Vapi, Nikhil Gupta, Yeapp, Ideamart, hSenid Mobile Solutions (Pvt) Ltd, Webio, Africa's Talking, Jonathan Marashlian, Tim Panton, Matthew D. Smith, Ken Herron, Brian West, Matthew Hodgson, Robert Vis, Vladimir Beloborodov, Mark J Harvey 马克, Ameed Jamous, Rob Pickering, Daniel Gill, Mark Hay, Paul Sweeney, Dave Horton, Karim Zaki, jambonz

I disagree with this article: https://lnkd.in/dDAi3bZm. I honestly didn't use LangChain on it's early versions, but now I've started using it and I've seen many benefits: - Observabilty: tools like #LangSmith / Langfuse (YC W23) automatically recognize the classes from LangChain and produce nice looking visualizations - Ready-made tools: for eg prompt templating, output parsers.. I've spent some time baking my on list parser, I could've used the LangChain class and being able to see the list nicely formatted in LangSmith - Streaming: many LangChain classes (eg JsonOutputParser) support streaming output. I've built my own parser in a product and I didn't support streaming, now I'll migrate it to LangChain - Quick starts: using LangChain is easy to just join components together using LCEL to get started: https://lnkd.in/dZd-JS-F However, I agree sometimes is important to keep a look at what LC is doing, for eg. many classes inject custom rules in the prompt. I feel that going "bare bone" will save on tokens, but at the cost of maintenability. For one ecommerce store-generator I've recently built I even used a visual workflow builder like FlowiseAI (YC S23) for eg. to make it easier to maintain the prompts and logic

Just got back from UNPARSED and I wanted to share a big thanks to Kane Simms, Tom Hewitson, and the rest of the VUX World crew for putting on such a great conference! It was wonderful to be back in the conversational AI and voice world, reconnecting with so many brilliant minds. Attending, one thing became clearer than ever: flow charts and directed dialog, the long-standing norms for dialog design, just don’t cut it in the age of LLMs. For years, we’ve tried to hammer Intents into these structures, like jamming round pegs into square holes. But with LLMs, the peg shape is a complete mystery! Flow charts have been our go-to for ages—simple, structured, and great for predictable interactions. When intents came along, we forced them into this framework, often with mixed results. Now, LLMs are here, bringing a whole new level of complexity and flexibility that these old-school methods simply can't handle. It’s like trying to fit a dynamic, nuanced conversation into a rigid, predefined mold—frustrating and impractical. Some folks are looking to state machines to solve these issues. As one of the original authors of SCXML, I can attest to how its complexity ballooned, making it clear that it wasn't the right approach for practical dialog design. State machines, while theoretically sound, often become overwhelmingly complex and impractical for real-world applications. Admittedly, I haven’t delved deeply into the latest state machine-based systems, so take my reaction with a grain of salt—it could just be my PTSD from earlier experiences. So, where do we go from here? We need to break away from old molds and embrace new methods of dialog design that truly harness the power of LLMs. It’s a challenging journey, but one full of potential. Ready to explore these new frontiers with me? #AI #ConversationalAI #LLM #Innovation #Unparsed2024 #FlowChartsAreOldNews #LetsInnovate

Yesterday we had the #7 GenAI meetup 🤖 Sara Nóbrega and RJ Burnham gave two amazing talks on Information retrieval with Azure AI Search and the journey between speech recognition and real-time voice GenAI, respectively. Huge kudos to both of them 👏👏👏 We'll be sharing the recordings soon! Stay tuned. Thank you to the 40 people who joined us! 😊 The Lisbon GenAI Community meetups will be back in September. And thank you to our community partner MachineMinds . You rock 💪 #xgeeks #KIgroup #GenAImeetup #techcommunity

🚨 The GenAI meetup is next week! 🚨 🔗 Register in the link in the comments 👇 👉 Sneak peek into our agenda: 7:00 PM 🎤 "From Speech Recognition to Real-Time Voice GenAI: A Journey Through Innovation" by RJ Burnham 💡Dive into the technological advancements that have driven this progression, with a focus on the tech stacks and architectural considerations crucial for developing robust real-time voice GenAI solutions. 7:30 PM 🎤 The second talk will be announced soon! 8:00 - 9:30 PM Networking, Snacks, and Drinks! 🍻🍕🎉 #xgeeks #KIgroup #genAI #meetup