Join top executives in San Francisco July 11-12 to hear how leaders are integrating and optimizing AI investments for success. Learn more
Data Labeling Platform Datasaurus today unveiled a new feature that allows users to tag data and train their own custom ChatGPT model. This latest tool offers a user-friendly interface that allows technical and non-technical people alike to rate and rank language model responses, which are further transformed into actionable insights.
With OpenAI president Greg Brockman one of the first investors, the company announced that its new offering is a direct response to the growing significance of natural language processing (NLP), especially ChatGPT and large language models (LLM ).
Datasaur said that professionals from various industries are eager to leverage this technology effectively. However, the need for more clarity and standardized approaches to building and training custom models has posed ongoing challenges. Many people find it difficult to tune and improve the performance of the many open source models available.
In response to this changing landscape, the company aims to provide comprehensive support for users in assembling their training data.
“Our goal is to provide users with the highest quality training data and help remove unwanted biases from the resulting model through our new offerings, while inheriting powerful capabilities from the existing Datasaur platform,” Ivan Lee, CEO and founder of Datasaur told VentureBeat. . “Our platform supports all types of NLP, whether it’s ‘traditional’ models like entity mining and text classification or new ones like LLMs. The goal is to ensure that all NLP tagging can happen on one platform instead of using spreadsheets for one type and open source tools for another.”
Evaluation of the quality of LLM responses
Datasaur claims that its latest additions, Rating and Ranking, are the easiest to use model training tools on the market today.
With grading, human annotators can assess the quality of LLM outputs and determine whether responses meet specific quality criteria.
Ranking facilitates the human feedback reinforcement learning (RLHF) process.
In addition to its new features, the platform introduces a reviewer mode that allows data scientists to assign multiple annotators, thereby minimizing subjective bias. This mode makes it easier to identify and resolve discrepancies between annotators when dealing with specific questions, allowing data scientists to make the final judgement.
The platform’s Inter-Annotator Agreement (IAA) feature uses statistical calculations to assess the level of agreement or disagreement between annotators. This tool helps data scientists identify annotators who may require additional training and recognize those who demonstrate a natural aptitude for this type of work.
Furthermore, the platform presents the original document from which the LLM drew the information. This serves two purposes: to prevent potential misinterpretations and to provide transparency in demonstrating the process used by the LLM.
Streamlining wider adoption of large language models
Datasaur’s Lee said industry professionals may not consider OpenAI models as viable options due to factors such as compliance, data privacy or strategic considerations. Lee also pointed out that the current focus of LLMs on the English language prevents users around the world from fully benefiting from these technological advances.
“NLP has come a long way in the last decade, and one of our important goals at Datasaur is to help automate manual labor as much as possible,” said Lee. “Datasaur’s mission is to democratize access to NLP by allowing users to work in any language, be it French, Korean or Arabic. We want this offering to help everyone more easily train and develop LLMs for their purposes.
The company says its platform has the potential to reduce the time and expense associated with data labeling by 30% to 80%.
To automate data labeling, the platform uses a number of techniques. Use established open source models such as spaCy and NLTK to identify common entities. It also uses the weak supervision method for data programming, allowing engineers to create simple functions that automatically label specific entity types. For example, if a text contains keywords such as “pizza” or “hamburger”, the platform applies the “food” classification.
Additionally, the platform incorporates an integrated OpenAI API, which allows customers to request ChatGPT to tag their documents on their behalf. The company says this approach can achieve high levels of success, depending on the complexity of the task, while also opening new avenues for automation.
According to Lee, the platform’s RLHF feature represents one of the most effective ways to enhance the training capabilities of an LLM. This approach, she said, allows users to quickly and effortlessly evaluate a set of model outputs and identify the superior ones, eliminating manual intervention.
“Our platform allows the user to display various options and rank them alphabetically from best to worst. The simple drag-and-drop interface is easy for a non-technical user to use, and the resulting output includes every permutation of ranking preferences (e.g. 1 is better than 2, 1 is better than 3, 2 is better than 3 ) to make it readily consumable by the technical data scientist and reward model,” Lee explained.
A future of opportunities in NLP
Lee noted that NLP investment within the market is thriving and he anticipates a rapid evolution of LLM-based products.
He said there will be a surge in the development of applications prioritizing LLM technology in the coming years.
“The next interfaces won’t be a chatbox; it will be inserted directly into the applications we use on a daily basis, such as Gmail, Word, etc. ”he said. “Just as we learned to optimize our Google search queries (e.g. “Starbucks hours Saturday”), mainstream audiences will feel comfortable interfacing with applications through this natural language interface Datasaur aims to be ready to empower and support organizations in building such data models and workflows.”
VentureBeat’s mission it is to be a digital city square for technical decision makers to gain insights into transformative business technology and transactions. Discover our Briefings.