An artificial intelligence training image data set developed by decentralized AI solution provider OORT saw considerable success on Google’s platform Kaggle.
OORT’s Diverse Tools Kaggle data set listing was released in early April; since then, it has climbed to the first page in multiple categories. Kaggle is a Google-owned online platform for data science and machine learning competitions, learning and collaboration.
Ramkumar Subramaniam, core contributor at crypto AI project OpenLedger, told Cointelegraph that “a front-page Kaggle ranking is a strong social signal, indicating that the data set is engaging the right communities of data scientists, machine learning engineers and practitioners.“
Max Li, founder and CEO of OORT, told Cointelegraph that the firm “observed promising engagement metrics that validate the early demand and relevance” of its training data gathered through a decentralized model. He added:
“The organic interest from the community, including active usage and contributions — demonstrates how decentralized, community-driven data pipelines like OORT’s can achieve rapid distribution and engagement without relying on centralized intermediaries.“
Li also said that in the coming months, OORT plans to release multiple other data sets. Among those is an in-car voice commands data set, one for smart home voice commands and another one for deepfake videos meant to improve AI-powered media verification.
Related: AI agents are coming for DeFi — Wallets are the weakest link
First page in multiple categories
The data set in question was independently verified by Cointelegraph to have reached the first page in Kaggle’s General AI, Retail & Shopping, Manufacturing, and Engineering categories earlier this month. At the time of publication, it lost those positions following a possibly unrelated data set update on May 6 and another on May 14.
While recognizing the achievement, Subramaniam told Cointelegraph that “it’s not a definitive indicator of real-world adoption or enterprise-grade quality.” He said that what sets OORT’s data set apart “is not just the ranking, but the provenance and incentive layer behind the data set.” He explained:
“Unlike centralized vendors that may rely on opaque pipelines, a transparent, token-incentivized system offers traceability, community curation, and the potential for continuous improvement assuming the right governance is in place.“
Lex Sokolin, partner at AI venture capital firm Generative Ventures, said that while he does not think these results are hard to replicate, “it does show that crypto projects can use decentralized incentives to organize economically valuable activity.”
Related: Sweat wallet adds AI assistant, expands to multichain DeFi
High-quality AI training data: a scarce commodity
Data published by AI research firm Epoch AI estimates that human-generated text AI training data will be exhausted in 2028. The pressure is high enough that investors are now mediating deals giving rights to copyrighted materials to AI companies.
Reports concerning increasingly scarce AI training data and how it may limit growth in the space have been circulating for years. While synthetic (AI-generated) data is increasingly used with at least some degree of success, human data is still largely viewed as the better alternative, higher-quality data that leads to better AI models.
When it comes to images for AI training specifically, things are becoming increasingly complicated with artists sabotaging training efforts on purpose. Meant to protect their images from being used for AI training without permission, Nightshade allows users to “poison” their images and severely degrade model performance.
Subramaniam said, “We’re entering an era where high-quality image data will become increasingly scarce.” He also recognized that this scarcity is made more dire by the increasing popularity of image poisoning:
“With the rise of techniques like image cloaking and adversarial watermarking to poison AI training, open-source datasets face a dual challenge: quantity and trust.”
In this situation, Subramaniam said that verifiable and community-sourced incentivized data sets are “more valuable than ever.” According to him, such projects “can become not just alternatives, but pillars of AI alignment and provenance in the data economy.“
Magazine: AI Eye: AI’s trained on AI content go MAD, is Threads a loss leader for AI data?
