Personalized AI assistant Coco lifts user-understanding proxy from 0.199 to 0.515 over 7 days
EchoShao8899 · x · 2026-11-17
The author shares results from a real-user study of Coco, a personalized AI assistant:
- Across 10 real users with varied backgrounds and AI familiarity, a macro-F1 proxy for how well the assistant understands the user rose from 0.199 to 0.515 over 7 days;
- Learned preferences evolved from broad cues to workflow-specific decisions;
- A notable case: on day 3, a non-technical participant reading technical material was offered on-the-fly explanations of unfamiliar terms plus comprehension checks — and kept using that support;
- The author's takeaway: personalization helps people discover new uses for AI, and the priority is scaling up the population that can benefit from it.
More from Models
- New MKA Benchmark: 200 Expert-Level Multimodal Tasks Across 83 Occupations — osanseviero · 2026-11-17
- OpenAI Brings in Apollo, METR, Transluce and Goodfire for Embedded Safety Assessments — OpenAI · 2026-11-17
- MKA Benchmark: Gemini 4 Carbon Scores 54.3% at $5.63 Per Task vs Claude's $35.28 — _philschmid · 2026-11-17
- Gemini Decisions API Judges 1,000 Questions Over 256K Tokens With Zero Output Token Billing — _philschmid · 2026-11-17
- RWKV-8 RPU preview hits 12,500+ tokens/s real-time inference — kastnerkyle · 2026-11-17
- Gemini 4 Helix spotted in Google Antigravity, hinting at another Gemini 4 variant — feblet · 2026-11-16