- With its new Data-centric Foundation Model Development capabilities, Snorkel AI aims to assist the users of the company’s Snorkel Flow platform to adapt to foundation models for Enterprise Use Cases.
- A second tool offered by Snorkel is the Foundation Model Warm Start feature, which utilizes an existing foundation model to assist with data labeling.
Data labeling is a time-consuming and complex facet of contemporary Machine Learning (ML) processes.
Data labeling can uncover foundation models’ enterprise potential. GPT-3 and DALL-E are helpful for creating text and graphics, but they need more context for corporate use cases. Tuning and training a foundation model warrants labeled data.
What if a foundation model could kick-start a data labeling operation to fine-tune smaller models for niche enterprise use cases? Snorkel AI claims to tackle this problem.
Alex Ratner, CEO and co-founder at Snorkel AI, said, “It’s one thing if you’re trying to do creative, generative tasks where you’re generating some copy text or some creative images, but there’s a big gulf between that and a complex production use case where you need to perform at a high accuracy bar for very specialized data and tasks,”
To address this issue, snorkel AI has published a preview of its new Data-centric Foundation Model Development capabilities. The objective is to assist Snorkel Flow platform customers in adapting foundation models for enterprise use cases. Ratner states that Snorkel’s primary research and concepts focus on identifying more efficient ways to label data to train models or refine them.
Building a new foundation for enterprise AI
Other manufacturers are also attempting to provide technology that facilitates foundation models’ refinement. Nvidia, for example, introduced its NeMo LLM (large language model) Service in September.
One of the fundamental features of the Nvidia service is the ability for customers to train huge models for specific use cases using the prompt learning technique. Using the prompt learning methodology, a companion model can add context to a pre-trained LLM, utilizing the prompt token.
The company’s Enterprise Foundation Model Management Suite, with the Foundation Model Prompt Builder feature, is known to make use of prompts. However, Ratner noted that prompts are only one component of a more comprehensive set of tools required to enhance foundation models for corporate use cases.
A second tool offered by Snorkel is the Foundation Model Warm Start feature, which utilizes an existing foundation model to assist with data labeling.
Ratner said, “So basically, when you upload a dataset to label in Snorkel Flow, you can now just get a push-button kind of first-pass auto labeling using the power of foundation models.”
Ratner remarked that Warm Start is not a solution for all data labeling, but it will be considered as the “low-hanging fruit.” Users are likely to utilize Warm Start with the prompt builder and the Foundation Model Fine-Tuning function of Snorkel to improve models. The fine-tuning feature allows businesses to convert the foundation model into a domain-specific training set.
An insight into generative vs. predictive AI enterprise use cases
Snorkel AI aims to apply foundation models for actual enterprise use cases.
Ratner stated that people are likely more familiar with generative AI that uses foundation models today, for better or worse. He characterized generative models as distinct from predictive artificial intelligence (AI) models, which help to forecast a result and are widely employed in businesses today.
Ratner related an anecdote in which he attempted to generate some Snorkel AI logos using Stable Diffusion because “..it was a ton of fun.” The actual corporate logo was an octopus wearing a snorkel submerged in the water, which he said he had to go through approximately 30 different designs to achieve!
Ratner said, “I guess that’s too odd of a nonsensical image, but I got some pretty cool logos after about 30 samples as a generative, creative human loop process. If you think of it from a predictive automation perspective, though, 30 tries to get one successful outcome is a 3.3% hit rate, and you can never ship something with that poor a result.”
Pixability, an online video advertising optimization provider, is one of Snorkel’s customers. Ratner explained that Pixability classifies millions of data points from YouTube videos for Machine Learning. Using the foundation model capabilities of Snorkel Flow, the classification can be completed quickly and with an accuracy of over 90%.
Ratner stated that a large U.S. bank that is a customer of Snorkel was able to improve the accuracy of text extraction from complex legal documents by implementing the foundation model strategy.
Ratner said, “We see this technology applying to the whole universe of applications where you’re trying to tag, classify, extract or label something at very high accuracy for some kind of predictive-automation task across text, PDF, image and video. We think it will accelerate all the use cases that we currently support and add new ones that wouldn’t have been feasible with our existing approaches before, so we’re quite excited.”