Cleanlab

One of the largest financial institutions in the world, BBVA, uses Cleanlab to improve their categorization of all financial transactions. Results achieved *without having to change their current model*: ➡️ Reduced labeling effort by 98% ➡️ Improved model accuracy by 28% This is the power of #DataCentricAI tools that provide automation to improve your data: Your existing (and future) models improve immediately with better data! Start practicing automated data improvement: https://cleanlab.ai/studio

In our latest blog, we walk through the benefits of reframing time-series data as a classification problem. This approach simultaneously leverages enhanced performance, flexibility, and interpretability while access to a variety of models like random forests and neural networks can capture complex patterns more effectively. Using a popular energy consumption dataset, we benchmark Cleanlab Studio AutoML alongside Prophet and Gradient Boosting - Cleanlab Studio's AutoML reached 94.61% accuracy, far outperforming other methods. Not only does Cleanlab Studio AutoML achieve superior results with minimal effort through automated modeling training, hyperparameter tuning and predictor selection, it simplifies deployment for production-level models, enabling real results in minutes. Accelerate development time and enhance forecast accuracy with Cleanlab Studio. Read the full blog and sign up today to see the difference it can make in your projects. https://lnkd.in/gKptsgBN

Give it a shot - sign up for Cleanlab Studio today. https://app.cleanlab.ai/

The problem is that most companies are investing 80% of their AI budget into models and only 20% into data. Here’s what needs to change to unlock AI value. Shift investment from gathering data to curating data. Curation builds a dataset for model consumers vs. human consumers. The cost of model training drops because reliable use case support is delivered with less data and less complex models. Shift investment from engineering data pipelines to engineering data-generating processes. Moving data from one place to another creates no value, while each new dataset makes the business more valuable. Data creates more AI opportunities. Unique datasets are the primary competitive advantage and AI moat. Models are only best-in-class for a few months. GPT-4o was upstaged by Gemini 1.5, which was just surpassed by Claude 3.5. The investment required to win on massive models is much too high for enterprise business models to support. Follow me here and click the link under my name to learn more about how to deliver value-centric AI. #AI #Data #AIStrategy #DataQuality

Is “data-driven with a disclaimer” an acceptable future for AI applications? Tomasz Tunguz posed the same question in one of his latest newsletters highlighting the implicit bias consumers feel toward their AI products. Tomasz asserts that the totally reasonable expectations we have for SaaS products to be both safe and accurate for enterprise doesn’t apply to the AI era—at least not yet. With every AI software sneaking disclaimers into their products (“Gemini may display inaccurate info…double-check its responses” or, “ChatGPT can make mistakes…check important info.”), we’ve all but accepted the reality that we can’t totally trust AI applications. And if we can’t trust them, we can’t fully embrace them. “We suffer from a cognitive bias: work performed by a human is likely more trustworthy because we understand the biases & the limitations. AIs are a Schrodinger’s cat stuffed in a black box. We don’t comprehend how the box works (yet), nor can we believe our eyes if the feline is dead or alive when we see it," says Tomasz. The more important our work is, the more confident we all need to be. Even human error rates are too much for the most financially—and socially—critical data use cases. Self-driving cars. Navigation systems. Insurance claims. News summaries. AI trust requires trustworthy AI. And trustworthy AI requires trustworthy data. How is your team going beyond the status quo to meet the real-time data quality demands of generative applications? Let me know in the comments!

Calling all AI enthusiasts and data science professionals in SF! Don't miss out on our Sake Social event happening on July 17th with Cleanlab and Open Data Science Conference (ODSC). Join us for a night of networking, learning, and curated sake 🍶✨. We are excited to have Curtis Northcutt deliver a data curation keynote that will surely spark interesting conversations around AI. This is a great opportunity to connect with like-minded individuals, expand your knowledge, and have a great time. RSVP now to secure your spot: https://bit.ly/4ex4d8L. See you there!

Don’t let messy docs run you RAG-ged! Cleanlab Studio now directly supports heterogenous document collections composed of files of the following types: doc, docx, pdf, ppt, pptx, csv, xls, xlsx. It’s an incredible way to organize your docs within a RAG system. Auto-detect issues across your heterogeneous documents and auto-label/tag them as well. Use a no-code interface to quickly get your docs ready for RAG use. ⚡ Read the whole blog: 👇 https://lnkd.in/g3zUUZQM

Yet another Foundation Model announcement that highlights the importance of data curation software. This time from Apple: "We find that data quality is essential to model success, so we utilize a hybrid data strategy in our training pipeline, incorporating both human-annotated and synthetic data, and conduct thorough data curation and filtering procedures." https://lnkd.in/gJgjCzzG

New tutorial - LLM Evaluation: Comparing Four Methods to Automatically Detect Errors Large Language Models (LLMs) are revolutionizing many industries, but they come with a significant challenge: hallucinations. These unintended outputs can lead to misleading or even harmful information. Addressing these errors effectively requires innovative approaches, especially when manual supervision is not feasible. In this tutorial, we explore four cutting-edge techniques for automated LLM error detection. Using a practical example with Shopify app store reviews, the tutorial will walk you through using each method to help improve the reliability of LLM outputs: 🔍 Token Probability Analysis: Assessing token confidence scores to identify potential inaccuracies. 🤖 LLM-as-Judge: Utilizing a secondary LLM to evaluate the accuracy of the primary model's labels. 🔄 Self-Consistency: Running multiple inferences with varied prompts to check for consistent results. 🛡 Confident Learning with Cleanlab: Leveraging auxiliary classification models to detect label inconsistencies. Each technique has its own strengths and trade-offs. Dive into the full tutorial to see detailed results and insights on choosing the best approach for your projects. 📖 You can try the full tutorial here: https://lnkd.in/dnias5e7 #LLM #AI #MachineLearning #ErrorDetection #DataScience #Automation #AIResearch #TechInnovation

The latest Scale Zeitgeist AI Readiness Report (2024) surveys 1800 ML practitioners across the US, from industries like Software/IT, Finance/Insurance, Business Service, Government, ... When asked to list "Top challenges in preparing high-quality training data for AI models", the #1 response was "Labeling quality", with "Curating data" ranking #4. The below image shows how the biggest challenges faced by AI Readiness respondents evolved over previous surveys, with Data Quality steadily growing in importance. At Cleanlab, we've long known how costly resolving data/label quality can be. It's vital for reliable AI, yet nobody wants to deal with this labor-intensive work. That's why we provide AI that can automatically find/fix dataset issues, as well as smart interfaces to quickly curate your {text, image, tabular} data.