NASA’s new dark energy space telescope can also detect killer asteroids
technologyreview.com·2h ago
Cloud & Infrastructure
The New Stack·27d ago·5 min read
Meta says it caught OpenAI. One thing is missing.
TL;DR
Meta claims its AI model, Watermelon, has achieved parity with OpenAI's GPT-5.5 on key benchmarks. However, the lack of specific benchmark data raises questions about the validity of this claim.
✦ Why It Matters
Engineers should advocate for transparent benchmarking practices to validate AI model claims effectively.
Key Takeaways
- Verify claims with public benchmarks before accepting them.
- Stay updated on Meta's AI developments for competitive insights.
- Consider the computational costs of AI models in your projects.
- Evaluate the credibility of internal claims from tech companies.
Full Summary
Meta's superintelligence chief, Alexandr Wang, announced that their AI model, Watermelon, has achieved parity with OpenAI's GPT-5.5 on unspecified benchmarks. However, this claim is based on internal discussions and lacks publicly available data for validation.
Meta has not provided a model card, evaluation metrics, or a release date for Watermelon, which raises questions about the credibility of the claim. Wang noted that Watermelon requires significantly more computational resources than its predecessor, Muse Spark, to reach this level of performance.
Meanwhile, OpenAI has already advanced to GPT-5.6, suggesting that Meta's progress may be lagging. Until verifiable benchmarks are released, the assertion remains unsubstantiated, highlighting ongoing challenges in Meta's AI development efforts.