Robert Chandler

💥 Finnt is now publicly available! 🌎 Building an AI-native software is a whole new beast - actually that’s why there’s so many great demos but so few amazing experiences in production. The nature of Generative AI makes it difficult to reproduce identical, deterministic outcomes. Where a traditional feature would have two statuses “functional” vs. “bugged”; in the GenAI world, a slightly different prompt or setup can drastically change the output. But this is also what makes AI so powerful and why we are so excited to be building in this space. 🧠 As a reminder, Finnt specializes in extracting crucial data from large, unstructured, and complex documents — including OMs, PPMs, 10-Q/10-K, as well as prospectuses, reports, leases, loans, etc. It then integrates these findings into AI-generated and templated memos, enriching them with real-time market data. 🔑 Key facts: - Ingest any large, complex, and unstructured files - Design custom frameworks tailored towards your specific use cases - Select your preferred LLMs, from closed to open-source options - Customize the goals, structures, and formats of your output reports - Augment your reports with third-party sources and live market data - Ask anything to your files and transform responses into insights - Seamlessly integrate Finnt into your existing workflows 👉 Check the comments to get the link to signup. PS: you might need to get a custom onboarding but we are nice people 😅

179

47 Comments

Today we're adding native AI support in Supabase Edge Functions, including: ◆ Embedding models ◆ Large language models (powered by Ollama) We've removed the serverless cold-boot by placing the models inside the edge runtime and we're rolling out a GPU-powered sidecar. My favourite thing about this setup: it runs locally; it runs on self hosted; it runs on the platform. Edge Functions can act like background workers for your Postgres database, offloading heavy compute, like generating vectors/embeddings to store with pgvector. Check out the demo video to see it in action, and read more in our latest blog post: https://lnkd.in/etJb8jFd #ai #databases #postgres #pgvector

148

7 Comments

“... native solutions built entirely around vectors will provide the ‘speed, memory safety, and scale’ needed as vector data explodes.”  “... we do advanced vector search in the best way possible.” At Epsilla (YC S23), we share the same vision and approach. Thus we have been leading the charge with our unique innovations in a 10x faster, cheaper, and better vector search engine, as the built-in foundation for our no-code RAG as a Service Platform. You can experience the accuracy and relevancy from our advanced technologies via a number of high quality AI-powered smart search and chatbot applications below: 1. Smart Search on 180+ RAG Research Papers (https://lnkd.in/eMPdVae5): This smart search app leverages our RAG platform and vector database technology to help user navigate through the massive 180+ papers with ease and keep up with the latest and greatest advanced research on this important topic. . 2. AI Assistant Chatbot for Knowledge Graphs (https://lnkd.in/eeBTFe_Q): Inspired by insights from Mike Dillinger's posts, this AI assistant chatbot leverages our RAG platform and vector database technology to generate highly relevant, nuanced and contextually aware responses to help user stay ahead of the game on topics like LLM, knowledge graph, and GenAI. 3. AI Assistant Chatbot for Taylor Swift (https://lnkd.in/e_m6dB4P): Powered by our RAG platform and vector database technology, this interactive AI-powered chatbot can answer any questions about Taylor Swift, one of the most impactful artists and successful business women in the world. These high quality AI applications underscore the fact that  Epsilla's vector database technology is not just keeping up with the demands of generative AI, but also setting new standards for what is possible. In addition, our no-code RAG platform allows AI app builders to develop and deploy high quality apps like these within hours. Join us as we continue to innovate and lead in this exciting era of AI-driven applications! (https://epsilla.com) #VectorDatabase #RAG #Chatbots #GenerativeAI #NoCodePlatforms

31

8 Comments

Diving into RAG (Retrieval-Augmented Generation) for LLMs - this was an eye-opener. Crafting the initial RAG application for Consolto? Simple. Scaling to a production-level solution? That’s where the real work begins. Here’s what we learned: - Initial setup vs. production-ready: getting started was quite straightforward with a few lines of code (found in probably a few thousands of sources across the net) but fine-tuning for production demands time, effort, and expertise. - Diverse approaches: every step in the RAG process -- tokenization, embedding functions, VectorDBs, evaluation, pre-processing, LLM completion etc.. -- offers multiple methods. Each approach shines for specific uses. - Balancing trade-offs: navigating trade-offs between cost, quality, speed, and security takes time & effort. There’s no universal solution yet, but the opportunity here is immense. The landscape is ripe for a go-to RAG solution to emerge... anyone interested in playing with the idea? (in the image - building the AI Knowledge-base from Consolto's dashboard, out. Link to register to the beta in the comments) #RAG #LLM #AI #MachineLearning #DataScience #AITrends

14

5 Comments

🚨big news from Emperia 🚨 Virtual worlds are notoriously hard to build - especially when it comes to running a high-fidelity experience on the web. They require a lot of asset optimisation, compromising on quality. Custom coding for interactions and commercial functionalities. And large teams involved in deploying each experience, which often takes months to create and deliver. Over the last 5 years, we’ve solved all those problems for our own 3D teams. We’ve built the underlying technology that enabled Emperia design, build and publish virtual experiences at scale. We always focused on visual quality and performance, and built the platform with the end user in mind. Today we’re officially opening our tools up for the creative community to build on Emperia platform. Built on UE5, Creator Tools enable: ✨Render and publish cross-device virtual experiences straight from Unreal Engine with no asset optimisation required ✨Plug-and-play UI, enabling any 3D artist to create a fully functional experience with no dev support required ✨ Procedural generation tools and lighting presets to create stunning environments in minutes ✨Commerce modules to support e-comm integration for sales & marketing experiences ✨Customizable interactions from gamification to token-gated rooms with an SDK to expand functionality even further ��and so much more! Shoutout to my Emperia team for making this happen, and our partners at Sawhorse Productions, Cominted Labs, Polycount.io, virtuall and many others who helped us refine our creator platform in beta over the past few months 🙏🏻 Simonas Holcmann Emma Preti Charlie Adams Jadir Lino Tomas Tamulenas Tomas Gavelis Zunaib Imtiaz Stephen Cherla Shahzaib Asim Marcelina Slabuszewska #UE5 #EmperiaCreatorTools #builtwithEmperia #3D #unrealengine #AWE2024

199

18 Comments

Can you type at 500 words per second? Guess not but Groq, a well-funded Silicon Valley startup, has been developing a novel processor architecture optimized for the matrix multiplication operations that are the computational heart of deep learning. Lets nerd talk: that’s 800 tokens per second on LLaMA 3. Groq's architecture is a significant departure from the designs used by Nvidia and other established chip makers. Instead of adapting general-purpose processors for AI, Groq has built its Tensor Streaming Processor to accelerate the specific computational patterns of deep learning. Here’s another trillion dollar valuation coming up at some point! In the picture below, it does look like the DeLorean so the llama is travelling at 88 mph #SpeedOfThought • #GrogInnovation • #AIProcessorRevolution • #DeepLearningTech • #TensorStreaming • #TechDisruption • #SiliconValleyStartup • #FutureOfAl • #MachineLearningSpeed • #TrillionDollarTech

3

1 Comment

🤩 The new DeepLearning.AI course is, in one word, AMAZING! It's so in-depth about multi-modal models (in this course, the team uses Google's Gemini and Weaviate as the AI-native vector database)! 🚢 so proud of the collaboration between all the teams involved in shipping this course!

34

3 Comments

This is a good article on using LLMs in the real world. It covers many aspects that we learned at Nurish over the last few months and more. Using LLMs is an inexact science, but has its own needs in processes and tools. While most of the attention has been on the LLMS, their capabilities, and RAG, there are mundane issues that need to be taken care of, issues where the solutions for pre-LLM development are not that well-suited. These include version management, integration, testing, data management, quality assessment, and more. This article covers many of those practical aspects. Eugene Yan, Bryan Bischof, Charles Frye & more https://applied-llms.org/

9

This clip pretty much sums up our take on AI at Mercoa https://lnkd.in/eNQuamSu Models from OpenAI and Anthropic are improving so quickly, and I don't think that will stop anytime soon. Our OCR engine was originally built on GPT-3 with a single prompt. Reaping the benefits from 3 to 3.5 to 4 to 4o has been a one line change. We are now rolling out more AI features based on this principle, and are able to do so extremely quickly. Talk about ROI!

11

Earn royalties for automating troubleshooting steps? See an AI Assistant use your steps to respond to alerts and questions? If you can be in San Francisco on Sat June 22nd, sign up for SRE Day and use discount code KYLE20 so we can provision you a dev environment for our workshop. Link in the comments. I'm also taking volunteers for anyone who wants to race Eager Edgar, one of our AI Assistants, troubleshooting popular Kubernetes-based apps... #sre #devops #platformengineering #whatdoirunwhen

76

10 Comments

🚀 Today I'm excited to launch the V2 of Openlayer: a Vercel-like workflow for AI evals that actually makes sense. ⬇️ Most of us get how crucial AI evals are now. The thing is, almost all the eval platforms we've seen are clunky and don't have any product cohesion. There's too much manual setup and adaptation needed, which breaks developers' workflows. That's why we're releasing a simpler workflow. If you're using GitHub, you only need to add two files to the repo (one config + one script). Then, connect your repo to Openlayer and define must-pass tests for your AI system. Once integrated, every commit triggers these tests automatically on Openlayer, ensuring continuous evaluation without extra effort. We offer 100+ tests (and are always adding more), including custom tests. We're language-agnostic, and you can customize the workflow using our CLI and REST API. As a final note, you can leverage the same setup to monitor your live AI systems after you deploy them. It's just a matter of setting some env vars in your staging/prod environments, and your Openlayer tests will run on top of your live data and send alerts if they start failing. Easiest way to get started is by heading to our docs: https://lnkd.in/d8JfaVsy

59

2 Comments

Now that our open source development project, for TGN VCS equipment for use in any modeler, is well underway and starting to show results, I’m thinking of ideas for first use to show off what it can do. Why not show TGN at work on an icon? We could show TGN VCS equipment in the Farnsworth House to really TELL THE STORY of how this building is put together. The virtual tour at the link gives a nice impression. But the medium (the virtual tour, as with any kind of model) has limits. The understanding developed remains kind of stuck at surface skimming level. Of course there have been countless attempts made at addressing that, attempts to dig deeper. This one is a great effort: https://lnkd.in/dVDtDMnY Look at all the information at that website, which is dated 1994, documenting the house, various kinds of description, portions modeled in 3D, images of the original construction drawings, videos, various analytics, photos… It’s a useful archive. I’m grateful to those who made it. Their names appear on a Credits page at the site. And no doubt, many have modeled this house since then. How many students have made it their project to 3D model the Farnsworth House, in Sketchup or Rhino or whatever favorite modeler? But take ALL OF THAT as a given and ask yourself, let’s say you were going to replicate Míes van der Rohe and build one of these, a Farnsworth house; sure, copy a modernist masterpiece, verbatim. Why not, in the spirit of ‘AI’! But I digress. Let’s say, you’re going to build something different but inspired by this house. How well do you understand this thing, what this house IS, how it’s actually put together, step by step, idea by idea, system by system, piece by piece? So you can learn from that and use the concepts on a different project, and differently.. Look at the drawings here: https://lnkd.in/dbRXdgcm Do you really understand where those drawings are, in context, and therefore what they mean? What about the 3D model? Take the best one you can find. Do you understand how this house is built by flying around in the model? Do you have any idea whether or not anything that matters is missing anywhere in the model? Where? And how can you even look and see how things are put together? The state of the art, of software supporting this, is just a joke. Still, in 2024. It’s stupid. What has the industry been doing all this time? I still just can’t believe how counter-productive and self-defeating the whole situation is. Come on now, let’s fix this once and for all. Set this industry on a new path that makes sense. Anyone interested in joining our open source project developing TGN VCS equipment for use in ALL modeling platforms, message me. The TGN VCS development outline and spec: https://lnkd.in/dUMj8Eqa

7

1 Comment

Having great documentation is one of our absolute top priorities at Cillers. One of the primary benefits of running hackathons is that we get really great feedback on how our documentation can be improved. Based on feedback from our AI Hackathon at Google last week, I've done a complete rewrite of our documentation. You can see it here: https://docs.cillers.com/ Would love to hear your thoughts and feedback!

21

3 Comments

Most AI bug fixing: LLM crawls a repro and hunts for bugs to fix. But Jam also gets the visual playback of a bug, and that's really useful signal for an AI bug fixing agent.

5

Tons of great points in this pod - one in particular sticks out to me: Beyond just improving employee productivity (improving the input), **AI enables software that delivers the work product output (delivering the outcome).** So, software companies can now look like a services business in terms of what they deliver for a buyer. #AI #b2b #SaaS

3

Making the case for digital twins using space data - or getting one step closer to thin footprint digital twins of our planet. TL:DR - apply data, knowledge, and visualisation at grassroots to solve problems at scale. https://lnkd.in/epuzdBn5

6

1 Comment

GPT-4-Turbo 🆚 GPT-4o Mikhail Burtsev recently shared some interesting findings on the performance of different GPT-4 models. According to his tweet from ICLR 2024, GPT-4 Turbo shows impressive results for simple question answering over documents up to 16K, while GPT-4o excels in shorter contexts up to 2K due to its lower cost and faster speed. For those dealing with mid-sized documents (6-10 pages), GPT-4 Turbo could be the way to go, offering higher quality outputs. On the flip side, the modifications in GPT-4o, aimed at increasing its speed, present a tradeoff that should be considered based on your specific needs. Before choosing your GPT-4 model, take a moment to consider which one aligns best with your tasks. 💡 Link to the original tweet by Mikhail Burtsev: https://lnkd.in/eEJ8695g #AI #GPT4 #MachineLearning #TechInsights #ICLR2024

3