In the news
FriendBench: Benchmarking Dyadic Familiarity Inference in Humans and Multimodal Large Language Models
arXiv cs.AI · Published · 3 min read
In 30 seconds
- What happened
- FriendBench benchmark tests whether AI models can infer if two people know each other from 20-second conversation clips across text, audio, and video.
- Why it matters
- Matters for engineers building social AI, emotion recognition systems, and multimodal models that need to understand human relationships from behavioral cues.
- Watch out
- Top models match human accuracy but use different reasoning: models bias toward "stranger" while humans stay balanced. Visual behavior helps humans more than models.
- language model
- prompt
- inference
- benchmark
The patterns behind this
Each one covers how the technique works, when it earns its cost, and where it breaks.
The Agent Architect
One pattern, one tradeoff, one production failure story. A short weekly briefing for people building agentic systems.
Weekly email, one-click unsubscribe. We only use your address to send the briefing.