Evaluation

152 briefings

• Token Routing Lifts Decode Throughput by 2.01–64.15x
• Causal Pretraining Makes Long Robot Context Pay Off
• Agentic Retrieval Gains 8.7 nDCG@10 Points Over Standard Retrieval With the Same Embedding Model

View topic →

Agent

150 briefings

• Artisan Posts Lowest Collision Rate Across 10+ Representative Models
• Causal Pretraining Makes Long Robot Context Pay Off
• 1T Cross-Region Relay-Tree Refit in 150 Seconds at 3% Change Rate

View topic →

Multimodal

150 briefings

• Artisan Posts Lowest Collision Rate Across 10+ Representative Models
• Token Routing Lifts Decode Throughput by 2.01–64.15x
• Pivot-SD Wins With 200 Problems; Queen Gains 915 Elo

View topic →

Training

129 briefings

• 1T Cross-Region Relay-Tree Refit in 150 Seconds at 3% Change Rate
• Agentic Retrieval Gains 8.7 nDCG@10 Points Over Standard Retrieval With the Same Embedding Model
• Pivot-SD Wins With 200 Problems; Queen Gains 915 Elo

View topic →

Safety

119 briefings

• Artisan Posts Lowest Collision Rate Across 10+ Representative Models
• Token Routing Lifts Decode Throughput by 2.01–64.15x
• Causal Pretraining Makes Long Robot Context Pay Off

View topic →

Image Gen

111 briefings

• Artisan Posts Lowest Collision Rate Across 10+ Representative Models
• Token Routing Lifts Decode Throughput by 2.01–64.15x
• TRM Defines Rubrics; SAKI Reaches 4.22× Throughput

View topic →

AI for Science

111 briefings

• Artisan Posts Lowest Collision Rate Across 10+ Representative Models
• Token Routing Lifts Decode Throughput by 2.01–64.15x
• Causal Pretraining Makes Long Robot Context Pay Off

View topic →

Efficiency

98 briefings

• Token Routing Lifts Decode Throughput by 2.01–64.15x
• Causal Pretraining Makes Long Robot Context Pay Off
• 1T Cross-Region Relay-Tree Refit in 150 Seconds at 3% Change Rate

View topic →

Architecture

97 briefings

• Artisan Posts Lowest Collision Rate Across 10+ Representative Models
• Token Routing Lifts Decode Throughput by 2.01–64.15x
• Agentic Retrieval Gains 8.7 nDCG@10 Points Over Standard Retrieval With the Same Embedding Model

View topic →

Video Gen

92 briefings

• Causal Pretraining Makes Long Robot Context Pay Off
• Pivot-SD Wins With 200 Problems; Queen Gains 915 Elo
• Chunked KV Compression Shows Retrieval Gaps of Up to 40 Percentage Points

View topic →

Robotics

89 briefings

• Artisan Posts Lowest Collision Rate Across 10+ Representative Models
• Token Routing Lifts Decode Throughput by 2.01–64.15x
• Causal Pretraining Makes Long Robot Context Pay Off

View topic →

Interpretability

77 briefings

• Causal Pretraining Makes Long Robot Context Pay Off
• 1T Cross-Region Relay-Tree Refit in 150 Seconds at 3% Change Rate
• Agentic Retrieval Gains 8.7 nDCG@10 Points Over Standard Retrieval With the Same Embedding Model

View topic →

Retrieval

76 briefings

• 1T Cross-Region Relay-Tree Refit in 150 Seconds at 3% Change Rate
• Mean OSWorld Task Score Rises From 33.0% to 40.8%
• AgentWorld's Best Model Succeeds Only 52% of the Time

View topic →

Reasoning

71 briefings

• Prism Trains 2.5× Faster and Improves Quality
• Chunked KV Compression Shows Retrieval Gaps of Up to 40 Percentage Points
• Qwen3.5-4B’s Nine-Benchmark Average Retention Reaches 82.3% at 5% Visual Tokens

View topic →

Code Intelligence

51 briefings

• Artisan Posts Lowest Collision Rate Across 10+ Representative Models
• Pivot-SD Wins With 200 Problems; Queen Gains 915 Elo
• Mean OSWorld Task Score Rises From 33.0% to 40.8%

View topic →