EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·26d ago·5 min read
Show HN: How Stale Is Your AI? Release age and training cutoff for 20 models
TL;DR
A new resource tracks the release and training cutoff dates for 20 AI models from 8 labs, revealing how current these models are. Only 9 out of 20 models have publicly available training cutoffs.
✦ Why It Matters
Check the training cutoffs of AI models you use to ensure they are current.
Key Takeaways
- Always verify the training cutoff date of AI models.
- Understand the gap between release and training cutoff.
- Use the provided resource to check model currency.
- Be cautious of AI responses regarding their own training data.
Full Summary
Two critical dates define the currency of an AI model: the release date, when the model is made available, and the training cutoff, which indicates when it last ingested new information. A new page lists these dates for 20 models from 8 different labs, with only 9 models having a publicly available training cutoff.
For instance, GPT-6 Astra was released on September 3, 2026, but its training data only includes information up to April 30, 2026, making it five months behind at launch. Users can directly ask models about their training cutoff, but caution is advised as models may provide inaccurate information.
This resource aims to clarify the gap between a model's release and its actual knowledge, emphasizing that models do not learn from web searches in real-time.