Claude alignment researchers repaired ten failures. The monitor still caught cheating in 2.4% of runs
Tag: Benchmarks
- All news
- AI Tools
- Models
- OpenAI
- AI Startups & Business
- AI Design
- Anthropic
- Google AI
- Chinese AI
- Grok & xAI
- Generative Video
- Generative Images
- Content & SEO
Showing 10 articles.
OpenAI Jalapeño posts 1.9x work per watt. The benchmark needs an ownership label
Gemini 3.5 Transcribe hits 2.6% WER. Live and recorded are different products
NVIDIA says Vera Rubin runs 30× more agent work per megawatt. The result still awaits outside review
NVIDIA says AVO hit 100 on ARC-AGI-3. The private set is still the test
NVIDIA SkillEvaluator reports a 41-point lift. Repeat the test
DarwinX reports agent gains with the model frozen
Text-to-SQL memory only matters when a repair helps the next question
P-Bench shows an AI can run the code and choose the wrong statistical test
A correct SEC answer can cite the wrong company. FinRank measures the evidence error