Current AI's 'Gap Map' Reveals Open Source AI Landscape, But Is It a Major shift?

Current AI, a non-profit backed by $400 million in funding, has launched its 'Gap Map', an ambitious attempt to index the state of open source AI. While the map itself offers an intriguing overview, the real story here is about data accessibility and the challenges of categorizing a rapidly evolving field.

The Gap Map v0.1 catalogs 421 products across 14 categories and three layers of the AI stack: model components, product/UX, and infrastructure. It breaks down into 266 software tools and libraries, 85 models, 50 datasets, and 20 hardware projects, all produced by 228 organizations. That's an impressive start, but it barely scratches the surface of the 24,400 additional 'artifacts' the project has identified as part of the 'long tail' of the open source AI ecosystem.

What sets this project apart isn't just the categorization, but the transparency of its data. Current AI has released the underlying information, 1,184 YAML files plus associated notebooks and scripts, under an MIT license on GitHub. This move towards openness is commendable, but it also highlights a key challenge: how do you meaningfully categorize and score such a vast and rapidly changing landscape?

The Gap Map's approach to this problem is telling. Those 24,400 uncategorized artifacts will carry no score until they are 'researched and cited.' This hints at the labor-intensive nature of the task and the potential for bias in what gets attention and what doesn't.

For developers and researchers, the real gold mine might not be the map itself, but the raw data. With 16,185 GitHub repositories tracked and made available as a CSV file, there's potential for deeper analysis and unexpected insights. Tools like Datasette Lite can be used to explore this data, opening up possibilities for the community to derive value beyond the initial categorization.

import datasette_lite
import pandas as pd

# Load the CSV data (assuming it's named 'os_ai_repos.csv')
df = pd.read_csv('os_ai_repos.csv')

# Use Datasette Lite to explore
datasette_lite.open(df)

However, the Gap Map's usefulness will ultimately depend on how frequently it's updated and how accurately it reflects the rapidly changing open source AI landscape. A static snapshot, no matter how comprehensive, quickly becomes outdated in this field.

The project's ambitious scope, from low-level model components to user-facing products and infrastructure, is both a strength and a potential weakness. While it provides a holistic view, it risks oversimplification in areas that deserve deeper, more nuanced analysis.

Ultimately, the Gap Map is a valuable starting point, but its true impact will be determined by how the AI community engages with and builds upon this data. Will it become a living, constantly updated resource, or a one-time snapshot that quickly loses relevance? The answer to that question will determine whether the Gap Map truly bridges the gap in our understanding of the open source AI landscape, or simply highlights how wide that gap remains.