Activeloop

Drug Discovery, meet Generative AI. Thrilled to unveil Flagship Pioneering's exciting achievement in the biotech sector: harnessing the power of Retrieval Augmented Generation (RAG) for faster drug discovery and improving retrieval accuracy by 18.5% with Activeloop Deep Lake. Flagship Pioneering is a biotechnology company that invents platforms and builds companies focused on making bigger leaps in human health and sustainability. The company’s work ranges from applications in human health, such as designing new therapeutic modalities or early cancer detection, to tackling challenges in climate and sustainability, such as finding more resilient forms of agriculture. Flagship, and its Pioneering Intelligence (PI) initiative, as well as Activeloop embarked on a collaboration to solve a challenge: efficiently answering complex scientific questions by searching through large-scale, multi-modal data without compromising accuracy or adding complexity? This is where Activeloop made a difference. Together, Pioneering Intelligence and Activeloop formed a research partnership to address these needs. PI developed systems to generate and evaluate “realistic” questions across a diverse range of biological topics that Flagship might pose during scientific exploration. Activeloop provided Deep Lake, the database for AI, and a capability called Deep Memory. Deep Memory enhances retrieval accuracy using a learnable index from labeled queries for specific RAG applications without affecting search time. 🚀 With Activeloop, Flagship Pioneering significantly improved their retrieval capabilities, with a 18% increase in accuracy compared to traditional methods, and streamlining drug discovery R&D process. 💡 Key Insights: 1️⃣ Efficiency & Accuracy: Deep Lake and Deep Memory enhance RAG applications in biotech with simpler data access and increased accuracy. 2️⃣ Multi-Layer Solutions to Solve Pressing Issues: The synergy between Flagship Pioneering, Activeloop, and the Intel Rise Program showcases how biotech's most pressing challenges can be addressed with AI-native data storage and computing (Intel XEON scalable processors). 3️⃣ Multi-Modality as a Key to Innovation: With plans to expand across data types, Flagship is leading industry innovation in how biologists interact with scientific data 🌟 Mark Kim summarized it best while talking about the collaboration and Deep Lake as a key building block for GenAI: In science, sometimes you have to rethink the basics to make progress. Flagship’s work with Activeloop has been all about that—getting back to the core of how we store and retrieve data for AI to speed up how we solve really tough scientific problems. This success wouldn't be possible without the support of Chris, Susan, and Arijit from the Intel Corporation & the thought leadership of Ian and Mark from Pioneering Intelligence, who are at the forefront of GenAI at Flagship Pioneering! Read the case study (in the comments) and watch the video below.

What a week for everyone in tech, huh? Ahead of today's AI at Meta Llama 3.1 release, there's a bunch of mindblowing news, including Llama 3.1 405B allegedly leaking on HF/4chan! - Llama-3.1 405B was made available for download on 4Chan. We dive deep into it in today's newsletter, but it's 15 Trillions tokens pre-trained (the number floating around for the OG GPT-4 was 13T); sports 128K context size, and might be partially paywalled! - Mistral AI released 3 new SLMs, that excel at maths, code, and multi-step logical reasoning. - OpenAI unveiled GPT-4o mini, a compact and cost-effective AI model for ChatGPT that outperforms leading small AI models on reasoning tasks while being 60% cheaper to operate than GPT-3.5 Turbo. - Our friends at Arcee.ai announced a Series A funding (congrats!), a powerful 'merged' SLM Arcee-Nova and two datasets for instruction following and agent training. - Salesforce released xLAM, a family of models for autonomous task planning and execution, with the 7B model achieving 88.24% on the BFCL function calling leaderboard. - Hugging Face SmolLM is a new line of efficient small language models designed for local devices, available in 135M, 360M, and 1.7B parameter sizes, outperforming similar-sized models like GPT-2 and MobileLM across various benchmarks. - FlashAttention-3 achieves up to 2× speedup in attention mechanisms using producer-consumer asynchrony and hardware-accelerated low-precision operations. - LMMs-Eval proposes a unified evaluation framework for multimodal AI, balancing task diversity, human alignment, and efficiency to enable standardized model comparisons. Read 20+ more news items across AI research, industry releases, and funding in our weekly AI newsletter and hit follow to stay tuned! https://lnkd.in/gQFAsHMU

What a week in AI: news from Microsoft, Anthropic, DeepMind, and a 3B model that can disrupt an entire document processing industry. - Everybody is talking about ColPali, a new retrieval model architecture that uses vision language models to directly embed page images, without relying on complex text extraction pipelines. Combined with a late interaction matching mechanism, ColPali largely outperforms modern document retrieval pipelines while being drastically faster and end-to-end trainable. Do we have another "SegmentAnything" moment ahead of us? - Microsoft resigned its observer seat on OpenAI's board amid antitrust scrutiny, while US lawmakers raised concerns about Microsoft's $1.5 billion investment in UAE-based AI firm G42 due to potential ties with China. - Anthropic announced fine-tuning capabilities for Claude 3 Haiku in Amazon Bedrock, allowing customization for specific business needs. - Google DeepMind's JEST promises to improve energy efficiency and model performance by using a smaller AI model to grade data quality. - A new method refines retrieved content before including it in the prompt for generation models, using meta-prompting to optimize instructions. - RTMW, a series of high-performance models for 2D/3D whole-body pose estimation, demonstrated strong performance while maintaining high inference efficiency. Read 20 more notable news items from last week in our newsletter below! https://lnkd.in/dSfMVwJy

What a week in AI! And the news just keeps coming, such as the release from EvolutionaryScale, a new AI research lab for biology, has raised $142 million in a seed round led by Nat Friedman, Daniel Gross, and Lux Capital (we'll cover that next week). - Looks like OpenAI's GPT-4 is in for some poetic justice from Anthropic's Claude Sonnet, with Sonnet 3.5 closing in on GPT-4o in LMSYS arena while being significantly cheaper (at $3 per MTok vs GPT-4o’s $5 per million of tokens and GPT-4's $30 per MTok). - xAI partnered with Dell Technologies and NVIDIA to build an AI factory, and Ilya Sutskever launched Safe Superintelligence (SSI) for AI safety. - Roblox is developing a 4D engine for interactive AI experiences that enhance realism and interactivity. - The MCT Self-Refine algorithm significantly improved success rates in solving Olympiad-level math problems (although we see some allegations on plagiarism put forward). - Kentauros AI, a new applied AI R&D lab, released the AgentSea open source agent platform. The toolkit is intended for agent creation and deployment, like a k8s style orchestrator called SurfKit that spins up agents, devices and tools. - A new MoA model outperformed GPT-4 at a fraction of the cost, achieving state-of-the-art results on multiple benchmarks.

Last week in AI was so packed with news, we had to vote which releases were the most thought-provoking! 🤯 Key developments: - LiveBench is a benchmark free from contamination and biases that top LLMs like GPT-4 and Claude 3 Opus weren’t able to score above 60% on. - Sakana AI found a way to find the best optimization algorithms for better LLM outputs using automated methods that wouldn’t be possible using traditional methods. - Musk is being sued by Tesla shareholders for starting xAI while he drops the lawsuit against OpenAI. - A team member from OpenAI’s Superalignment team published a controversial and thought-provoking two-hour-long piece on strategic considerations on the AGI race. - Apple unveiled on-device and foundation models, while NVIDIA introduced the Nemotron-4 340B to help generate high-quality synthetic data. Read 20+ more news items across investments, industry releases, and research in AI: https://lnkd.in/e6MmDxZd

Activeloop is Named a 2024 Gartner® Cool Vendor in Data Management. We're thrilled to announce Activeloop has been recognized by Gartner as a "cool vendor" in their latest report, Cool Vendors in Data Management: GenAI Disrupts Traditional Technologies! Gartner’s Cool Vendors reports identify "interesting, new, and innovative vendors, products, and services". This is an important milestone that underscores our company’s potential to disrupt the data management landscape with the Database for AI. This achievement reflects our unwavering commitment to innovation and excellence in the AI data management space. It's a testament to the tireless efforts of our entire team, alongside our valued collaborations with industry leaders like Flagship Pioneering, Bayer Radiology, Matterport, as well as technology giants like Intel Corporation, Amazon Web Services (AWS), Microsoft and more. I'm beyond excited to share this milestone! Many thanks to the entire Gartner team of analysts, and specifically Nina Showell, Aaron Rosenbaum, Ehtisham Zaidi, and Rick Greenwald for including us in this report! Learn more in the blogpost.

Real Estate, Meet Multi-Modal AI. We're excited to highlight an innovative use case from an Activeloop customer: leveraging multi-modal AI to transform the visualization and analysis of real estate properties. Matterport, a leader in 3D digital twins, has digitized more than 35 billion (!) square feet, making them one of the most prominent players in the domain. The company's Vision & Learning team partnered with Activeloop to streamline how they store complex data and retrieve it to train machine learning models at scale on opt-in data. Results? - 80% less time spent on data prep - Time to train on a new dataset reduced from hours to seconds Here are the challenges Matterport tackled with Activeloop: 1️⃣ Rapidly evolving vast data With its capacity to handle multimodal data, Deep Lake significantly streamlined the data handling process for Matterport's machine learning projects. 2️⃣ Lack of standardization Deep Lake provided a uniform, efficient storage format for Matterport's datasets, allowing stakeholders across teams to store data in a singular, ML-native format and abstracting away much of the boilerplate code required to set up a training pipeline for one project. 3️⃣ Multimodal support With its capacity to handle multimodal data, Deep Lake streamlined Matterport's machine learning projects' data handling process, allowing it to scale horizontally across data types without any additional legwork. By improving productivity, Activeloop allowed the Matterport team to focus their time on what actually matters - rapidly iterating on cutting-edge research and implementing it in Matterport's product lines. Alan Dolhasz, Manager of Machine Learning Development at Matterport, summarized it best: "Deep Lake knocked out about 80 percent of the data random work associated... It made working on more complex data no more complicated from a data management point of view. Whether I'm working on 10 million images with ten different modalities or a thousand images with just one modality, it's all the same from the perspective of the user of the system." I'm thrilled about this release! A huge thanks to Alan and his entire team for being an invaluable design partner, enabling us to achieve excellence as they develop AI products at the forefront of innovation. Our partnership has thrived over the years, evolving from an initial open-source community inquiry into a robust collaboration that has produced outstanding results! Read the complete case study (link in the comments).

Real Estate, Meet Multi-Modal AI. We're excited to highlight an innovative use case from an Activeloop customer: leveraging multi-modal AI to transform the visualization and analysis of real estate properties. Matterport, a leader in 3D digital twins, has digitized more than 35 billion (!) square feet, making them one of the most prominent players in the domain. The company's Vision & Learning team partnered with Activeloop to streamline how they store complex data and retrieve it to train machine learning models at scale on opt-in data. Results? - 80% less time spent on data prep - Time to train on a new dataset reduced from hours to seconds Here are the challenges Matterport tackled with Activeloop: 1️⃣ Rapidly evolving vast data With its capacity to handle multimodal data, Deep Lake significantly streamlined the data handling process for Matterport's machine learning projects. 2️⃣ Lack of standardization Deep Lake provided a uniform, efficient storage format for Matterport's datasets, allowing stakeholders across teams to store data in a singular, ML-native format and abstracting away much of the boilerplate code required to set up a training pipeline for one project. 3️⃣ Multimodal support With its capacity to handle multimodal data, Deep Lake streamlined Matterport's machine learning projects' data handling process, allowing it to scale horizontally across data types without any additional legwork. By improving productivity, Activeloop allowed the Matterport team to focus their time on what actually matters - rapidly iterating on cutting-edge research and implementing it in Matterport's product lines. Alan Dolhasz, Manager of Machine Learning Development at Matterport, summarized it best: "Deep Lake knocked out about 80 percent of the data random work associated... It made working on more complex data no more complicated from a data management point of view. Whether I'm working on 10 million images with ten different modalities or a thousand images with just one modality, it's all the same from the perspective of the user of the system." I'm thrilled about this release! A huge thanks to Alan and his entire team for being an invaluable design partner, enabling us to achieve excellence as they develop AI products at the forefront of innovation. Our partnership has thrived over the years, evolving from an initial open-source community inquiry into a robust collaboration that has produced outstanding results! Read the complete case study (link in the comments).

Just a couple of last spots left on the waitlist for todays packed event at LlamaIndex HQ, with speakers from Activeloop (Mikayel), Tryolabs (Diego), and LlamaIndex (Laurie) who talk about the latest advancements in GenAI and how to take them into production, moving beyond vanilla RAG applications. Final chance to attend before we close off the waitlist! https://lnkd.in/dMde_eTZ?

Just a couple of last spots left on the waitlist for todays packed event at LlamaIndex HQ, with speakers from Activeloop (Mikayel), Tryolabs (Diego), and LlamaIndex (Laurie) who talk about the latest advancements in GenAI and how to take them into production, moving beyond vanilla RAG applications. Final chance to attend before we close off the waitlist! https://lnkd.in/dMde_eTZ?