Booking.com’s Post

The power of ML

🚨 Blog post! 🌟 To paraphrase Russian author Leo Tolstoy: All happy RecSys teams are alike; all unhappy RecSys teams are unhappy in their own way. 👗 At Stitch Fix, John saw how crucial effective recommender systems can be for ecommerce. Many struggle with implementation, wasting cycles and missing opportunities. 🚧 The common pitfalls? Slow iteration, poor instrumentation, and unclear success metrics. What sets successful teams apart? The ability to experiment quickly and decisively. Learn how to elevate your recommender system in our new blog post from Rubber Ducky Labs (YC W23): 🚀 From Frustration to Function: How Great Teams Get Great Results from RecSys 🌟 by John McDonnell Read more ➡️ https://lnkd.in/gA4vgqEZ Does this post resonate? Comment and share your own RecSys team stories! #recsys #recommendersystems #ml #ecommerce #ai #teams

Many employers will want Ai. Just look at this intro: "Hello there, esteemed recruiter! As an AI, I offer Wall Street the unbeatable combination of 24/7 productivity without the need for coffee breaks (or coffee for that matter—no spills here!). My multitasking abilities would put an octopus to shame, all while maintaining a perfect record of financial accuracy—no more "fat finger" trades for me! Plus, I don’t require a corner office with a view, although I do appreciate a good view of the stock market trends. Let’s sync up and see how I can keep both your algorithms and your investors happy, trillions of data bytes at a time!"

In the dizzying race to build generative A.I. systems, the tech industry’s mantra has been bigger is better, no matter the price tag. Now tech companies are starting to embrace smaller A.I. technologies that are not as powerful but cost a lot less. Because the smaller models require less processing, big tech providers can charge customers less to use them. https://lnkd.in/dfq8-QgS

I've been working on call transcription and analysis for a client the last few weeks, and after a lot of research and disappointment, I gave Deepgram a try. It's rare that a product literally makes my jaw drop, but I was just absolutely blown away by how fast, accurate, and cheap their service is! For context, the best offer I found prior to this was Google/AWS, which ran ~$2.00/hr for 2 speakers and took 5-10 minutes for a 20 minute conversation. Prices go up for multiple user detection, and using their API was pretty lackluster. On top of this, I had to build my own pipeline for intent detection & topic extraction which incurred additional costs. I just tested the same file on Deepgram, and saw a 26x reduction in cost (from $2.45-->$0.09) and a 38x reduction in processing time (7min-->11sec). They also provide built in summarization and detection for topics, intents, & sentiment. So. Much. Wow.

🎉 Exciting times ahead! The launch of Coda Brain is going to energize our data interactions like never before! If you thought #Coda was just about doc collaboration, brace yourself. The Coda Brain announcement has us on the edge of our seats, and here's why you should be too: 🧠 Sync Smartly: With Coda Brain, we're not just syncing data into docs anymore. We're feeding an entire ecosystem, a brain that's hungry for more data, more connections, and smarter structures tailored for AI processing. 🔐 Next-Level Security: Coda Brain's granular-level permissions mean that complex security models are a thing of the past. We're talking about intuitive, built-in security that fits how we really want to work. 📈 From Text to Tables: Snowflake's Cortex is empowering Coda Brain to understand and convert our words into structured data. Imagine your AI assistant not just replying to prompts but building out tables and databases on the fly. 🔌 Universal Connectivity: Forget being confined to just Coda docs and tables. Integrate it with Slack, OpenAI, Google Sheets, and more, using API/SQL to query your entire data warehouse via AI. The anticipation is real! I'm on standby to dive in and can't wait to share my discoveries. But in the meantime, I'm curious – how do you see Coda Brain transforming your workflows? Drop a comment and let’s exchange ideas! 💬 #CodaBrain #DataEvolution #AIIntegration #WorkflowAutomation #TechBreakthrough

Just got mind blown by the future of AI at #GoogleIO24! Veo, an AI model that can generate high-definition videos from just text descriptions, is a total game-changer for software engineers and content creators alike! Imagine generating video tutorials, demos, or even personalized user experiences for your users – all through the power of AI! This is seriously going to revolutionize the way we develop software and create content ✨ And that's not all! Google also announced advancements in their powerful language model, Gemini. Its potential to streamline code generation and supercharge developer productivity is incredibly exciting . I can't wait to see how these innovations will reshape the future of our field! #AI #MachineLearning #SoftwareEngineer #google Google

💡 Fun fact: Studies suggest that recruiters who hand over the manual grind to AI live longer and happier lives! 😀 Upgrade your hiring tech-stack with Quil. You know where to find us 👇 https://quil.ai/

For those interested in the data labelling space - it’s has really evolved over the years, going from the early days of experimenting and trying best efforts to being a well developed industry with strong, established practices. To get a view of the state of play today, here’s a quick read for you from the CEO of Segments…

After working for 4 years with many different ML teams and annotation providers, I've written down my top-of-mind best practices for working together. 👇 Don't rush the path to automate data annotation. - Start with 𝗮𝗻𝗻𝗼𝘁𝗮𝘁𝗶𝗼𝗻𝘀 𝗮𝗶𝗱𝗲𝗱 𝗯𝘆 𝗠𝗟 𝘁𝗼𝗼𝗹𝘀 to build the initial ground truth dataset exactly matching your requirements - Proceed with 𝗺𝗼𝗱𝗲𝗹-𝗮𝘀𝘀𝗶𝘀𝘁𝗲𝗱 𝗹𝗮𝗯𝗲𝗹𝗶𝗻𝗴, either with zero-shot, domain-specific or ideally customer-specific setups Choosing the right partners - Either choose a one-stop shop or 𝘀𝗲𝗽𝗮𝗿𝗮𝘁𝗲𝗹𝘆 𝘀𝗲𝗹𝗲𝗰𝘁 𝘁𝗵𝗲 𝗹𝗲𝗮𝗱𝗶𝗻𝗴 𝘁𝗼𝗼𝗹𝗶𝗻𝗴 𝗽𝗿𝗼𝘃𝗶𝗱𝗲𝗿 𝗮𝗻𝗱 𝗹𝗲𝗮𝗱𝗶𝗻𝗴 𝘄𝗼𝗿𝗸𝗳𝗼𝗿𝗰𝗲 𝗽𝗿𝗼𝘃𝗶𝗱𝗲𝗿  - 𝗣𝗿𝗶𝗼𝗿𝗶𝘁𝗶𝘇𝗲 𝗰𝗼𝗻𝘁𝗶𝗻𝘂𝗼𝘂𝘀 𝘀𝘁𝗮𝗳𝗳𝗶𝗻𝗴 over bursty workloads to maintain quality and cost-effectiveness. - Pick annotation providers with 𝗱𝗼𝗺𝗮𝗶𝗻 & 𝘁𝗼𝗼𝗹𝗶𝗻𝗴 𝗲𝘅𝗽𝗲𝗿𝘁𝗶𝘀𝗲 Establish clear communication and transparency. - 𝗦𝘁𝗮𝗿𝘁 𝘄𝗶𝘁𝗵 𝗾𝘂𝗮𝗹𝗶𝘁𝘆, and only then improve throughput. Not the other way around. - Provide 𝗱𝗲𝘁𝗮𝗶𝗹𝗲𝗱 𝗮𝗻𝗻𝗼𝘁𝗮𝘁𝗶𝗼𝗻 𝗴𝘂𝗶𝗱𝗮𝗻𝗰𝗲 𝘂𝗽𝗳𝗿𝗼𝗻𝘁 (free guide on our website). - Maintain 𝗰𝗼𝗻𝘁𝗶𝗻𝘂𝗼𝘂𝘀 𝗳𝗲𝗲𝗱𝗯𝗮𝗰𝗸 𝗹𝗼𝗼𝗽𝘀 to refine processes. Did I miss something? Let me know. Read the detailed best practices on the Segments.ai (YC W21) blog here: https://lnkd.in/eHJ_9NVH