NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·1h ago
Cloud & Infrastructure
The New Stack·Jun 20·5 min read
Losing Fable made the best case yet for AI models you can run yourself
TL;DR
A significant issue in AI is that hosted models can be taken offline unexpectedly, limiting user control. Fable, a promising AI model, was quickly removed after its launch, highlighting the risks of reliance on hosted solutions.
✦ Why It Matters
Engineers and researchers should prioritize self-hosted AI models to maintain control and avoid service disruptions.
Key Takeaways
- Consider transitioning to open-weight models for greater control.
- Evaluate the cost-effectiveness of self-hosting AI models.
- Stay informed about government regulations affecting AI access.
- Test and qualify open-weight models against your existing workflows.
Full Summary
Fable, an AI model launched by Anthropic, was taken offline just three days after its release due to a U.S. government export-control directive. This incident underscores a critical lesson: access to a hosted model does not equate to ownership, as users can lose access without recourse.
In contrast, Z.ai's GLM-5.2 offers open weights that users can download and run on their own infrastructure, making it a more reliable option. The model has garnered attention for its performance, reportedly outperforming other models in coding tasks while being significantly cheaper to operate.
As the gap between frontier models and open-weight models narrows, the economic advantages of self-hosting become increasingly compelling, with predictions that most AI power users will self-host within a few years.