Open-Weight is Not Open-Source: Gary Marcus Slams Meta's Misleading Marketing
Gary Marcus · rss · 2026-08-10
Gary Marcus argues that media outlets like The New York Times are confusing Meta's new 'open-weight' model with true 'open-source'.
He emphasizes that genuine open-source provides complete transparency and customizability, allowing users to access and modify the entire pipeline, including training data and algorithms (e.g., AllenAI’s Olmo). In contrast, open-weight models only provide the final neural network parameters without revealing the raw data or preprocessing.
This distinction means developers cannot fundamentally alter the model or remove dataset biases, and scientists cannot investigate core AI mechanisms like regurgitation versus generalization. Marcus criticizes Meta for deliberately piggybacking on the positive PR of 'open-source' while withholding crucial details, urging the public to stop conflating the two terms.
More from Companies & People
- AI Code Flood Strains Reviews: Meta Diff Size Up 106%, System Near Collapse — rseroter · 2026-08-11
- OpenAI CFO Shares 5 Lessons for Building an AI-Native Finance Function — OpenAI News · 2026-08-11
- Report: SpaceX Could Close $60B Acquisition of AI Coding Startup Cursor Next Week — DimaZeniuk · 2026-08-11
- Meta's Open-Weight Muse Glimmer Model Hints at Zuckerberg’s Superintelligence Vision — TechCrunch AI · 2026-08-11
- Spotify Open-Sources Xirp: Agentic Dev Environment for Claude, Gemini, and Codex — omooretweets · 2026-08-11
- US Open Models to Outcompete Chinese Labs, Forcing Ecosystem Re-evaluation — qinzytech · 2026-08-11