Labelbox’s latest innovations: New features for coding, audio, talent acquisition, and more

We had an incredibly productive first month of the year, consistently delivering new features and capabilities week after week. Exciting new features improve our ability to train models on coding, enhance audio speech-to-text tasks, improve efficiency reviewing documents with automatic OCR, and more. Together, these updates focus on our primary mission to deliver high-quality datasets for training frontier models—delivered quickly, with high confidence, and with the flexibility to match your needs. 

In this blog we review the key enhancements to the Labelbox Platform, to Alignerr Connect, and to our cutting-edge AI interview tool, Zara. Read on for the details. 

Elevating the coding trainer experience

We're committed to providing the best possible tools to help software developers and programmers train AI models. Over the past month we’ve introduced a big improvement to our code editing capabilities in Labelbox. 

Code editor introduces Monaco for major update

Code editing within Labelbox has received a major upgrade (with even more coming in the near future so stay tuned). We've integrated the Monaco editor—the same powerful editor that drives Visual Studio Code—into the Labelbox Multimodal Chat editor. This new integration provides a number of key quality-of-life improvements when writing and reviewing code. 

The integrated editor brings improved syntax highlighting, line numbers, code folding, and even auto-completion for a smoother, more efficient coding experience. We've also enhanced the Code Runner with support for environment variables and multi-file project execution, and significantly improved the reliability of our Code Critic by leveraging Gemini’s structured output.

When editing and writing code in the editor, you'll now see an edit-in-code button at the top right next to any code block. By clicking on that, you get a focused code editing experience, again, powered by Monaco with all the core features found in preview. Along with the line numbering, syntax highlighting, and other expected features, you can now enjoy other key features like auto-completion. Once you're done editing code, you click ‘Save’ and are brought back to the markdown editor, where you can keep editing the regular text as well.

Improving labeling efficiency and accuracy

We're tackling the most time-consuming and error-prone labeling and data generation tasks head-on with powerful new automation features. Enhancements to the Labelbox Platform further improve our capabilities for audio tasks and accelerate tasks that require text extraction from PDFs and documents. 

AI-powered audio speech recognition

Extracting text from audio for speech-to-text, topic modeling, and text generation is now dramatically faster and more accurate. Our new auto-transcription feature within the audio editor, powered by the Whisper model, automatically transcribes audio segments with just a few clicks. This new automation eliminates the tedious manual transcription process, dramatically reducing the labeling time and minimizing the risk of human error.

To use the new feature, simply create an audio project and configure your ontology with a temporal text classification field. After adding an audio file, begin labeling and click on the temporal text classification. Select a frame in the timeline and the “Start Transcribing” button will appear. Click to start (and stop) the transcription. The text appears in the timeline and can be selected to automatically copied into your free-text box. 

Auto-extract text from PDFs and documents

Extracting text from PDFs for natural language processing (NLP) and text summarization projects is now a breeze. AI trainers can simply draw a bounding box around the desired text, and our ChatGPT-powered OCR engine automatically recognizes and extracts it. This eliminates manual copying and pasting, significantly accelerating the data preparation process.

To use the new feature, add a bounding box object to your ontology and toggle the “Automatic OCR” button. When you select that label in the Document editor and draw a bounding box, the text is automatically generated for you (as shown in the image below using the built in OCR capabilities).

Streamlining the AI talent acquisition

Finding the right AI expertise is crucial for success. We're making it easier than ever for you to identify and connect with top talent through Alignerr Connect. 

Alignerr Connect enhanced user interface

We've polished the Alignerr Connect user interface within the Labelbox Platform, adding new advanced filters and improving other flows to create a more seamless and intuitive experience. Finding the perfect Alignerr with the precise skills and experience you need is now simpler than ever.

Alignerr Connect public marketplace

To further connect you with the expertise you need, we've launched a public Alignerr Connect marketplace on our website—no app login required. This new resource showcases a curated selection of top-tier Alignerrs, highlighting their specializations, experience levels, and areas of expertise. Browse profiles, filter by skills like coding, STEM, or complex reasoning, and discover the perfect AI trainers for your next project.  

While a more comprehensive Alignerr Connect experience is available within the Labelbox Platform, this public marketplace offers a sample of the talent pool available. If you have a login for a paid Labelbox Account, then visit the Services tab in the Labelbox Platform to view the comprehensive set of available Alignerrs and advanced filters. And if not, then contact us to learn more about how our expert network can accelerate your AI initiatives.

Seamless Zara interviews

We're continuously refining Zara, our powerful AI-powered interview tool, to provide the most effective way to assess candidates. We've improved the connectivity to and overall responsiveness of Zara. These updates create a more natural and engaging experience for candidates, while backend enhancements provide more robust scoring for faster acceptance of top talent. In addition, further security and fraud prevention innovations have been added to further help identify top talent more effectively.

Looking ahead

These releases from the past four weeks represent just a fraction of the innovation happening at Labelbox. We're committed to continuously pushing the boundaries of what's possible in delivering the highest-quality data and the most talented team of AI trainers for developing frontier models. 

Stay tuned for more exciting updates in the weeks and months to come as we continue to empower our customers to build the next generation of AI models. We encourage you to explore these new features and discover how they can help you accelerate your AI initiatives.