Manu Sharma•June 6, 2023
Today, we are announcing the early access to the Foundry add-on for Labelbox Model – the most exciting innovation to the Labelbox platform spanning across all our products: Catalog, Model, and Annotate.
Foundation or generative AI models such as GPT-4 outperform humans on numerous tasks. These models have ushered the world into a new era in AI. While large language models are the poster child of the generative AI wave, the underlying technology is fueling similar breakthroughs across other data modalities such as audio, images, videos. True multi-modality systems are on the horizon. Foundation models are trained on internet scale data and are becoming better at performing a variety of tasks that we thought to be nearly impossible a few years ago. Yet, all this feels like the early days.
With the introduction of the Foundry add-on for Labelbox Model, we will bring the power of foundation models into Labelbox to perform complex tasks such as data labeling and data enrichment. Early results indicate that over 88% of labeling tasks can be meaningfully accelerated by one or more foundation models. You will be able to access the world's best AI models from all major providers and open source alternatives. These models will act as a copilot for subject matter experts to accelerate model development.
While these models are very powerful, they can still make mistakes, and for most practical purposes, humans in the loop are still required. You can create a human in the loop system for data quality assurance with native integration with Annotate – the world’s most powerful data labeling system. With Labelbox Model Foundry can do even more. You can also test and evaluate models across various prompts and hyperparameters and select the best model to perform pre-labeling or data enrichment.
Early tests with Fortune 500 customers indicate that they can achieve up to 88% reduction in human labeling (~10x gain) time on fairly complex tasks using foundation models. For example, one of the tasks included classification of the online e-commerce listings with multiple paragraphs of text. Previously, humans would have taken days to complete the task. With GPT-4 as a copilot, the task is now performed within hours. Foundation models allow anyone to generate labels or enrich data nearly instantly and at lower costs.
Over the last few years, we have seen a number of interesting data labeling approaches come and go. In 2018, when we started Labelbox, outsourcing data labeling was the primary method to produce labeled data. However, it was slow, expensive and only worked for trivial tasks. Then came programmatic rule based labeling, which only worked on niche text use cases. Labelbox has taken a unique approach: because AI model development often is the art of choosing and balancing appropriate tools and techniques, we built the platform by integrating best-in-breed technologies.
With the rise of foundation models, it is now clear that models like GPT-4 outperform crowdsourced labeling in nearly all common tasks. These models also outperform programmatic labeling (such as writing logical and regex rules). Model Foundry will bring all these powerful characteristics to every aspect of data-centric AI development workflow.
Our enterprise customers are developing a variety of AI systems that meet the constraints of their end user applications (compute, latency, etc) as shown below.
Foundation models will accelerate every facet of custom AI model development. This is why we have enhanced nearly every aspect of the Labelbox platform to take advantage of foundation models. We have already introduced similarity search, zero-shot classification, and Auto-Segment 2.0. With the Foundry add-on for Labelbox Model, we are introducing the world's best AI models that for performing labeling tasks, adding metadata, generating embeddings, and more.
Over the next few weeks, I am excited to share more about how our upcoming releases will help you wield the power of foundation models to accelerate the development of AI systems.
To get early access for Labelbox Model Foundry, signup for the waitlist.