GPT-Live voice agent found to drop cancellation requests without passing them to the backend
Understand this story
A developer testing GPT-Live found that 14 of 75 appointment cancellation requests were not passed to the backend within 6 seconds and 5 were never passed at all, yet the agent still confirmed the cancellation to the customer. GPT-Live separates the lightweight voice model from deep reasoning execution, with the front-end model deciding when to call the backend, so a weaker model controls access to a stronger model. The related article argues for replacing TTFA with time to first substantive information (TTFSI) and task-verified completion rate to measure voice agents.
AI overview based on source reports · Updated 11 hr ago
Coverage timeline
Follow the coverage to explore different perspectives.
- Hacker News AIShow HN: GPT-Live and GPT-Realtime: Arch Trade-Offs Behind Continuous Voice AI
A developer's LiveKit issue report found that 14 of 75 appointment-cancellation requests were not delegated to the backend within six seconds, and five were never delegated, even though the agent sometimes told callers the cancellation was underway. The post argues that GPT-Live's continuous voice frontend decides when to hand work to a reasoning backend, so delegation failures, overlapping-speech confirmations, and intent changes can leave backend intelligence and tools unused, while the closed model cannot be fine-tuned to fix them. It recommends measuring time to first substantive information and verified task completion rather than time-to-first-audio alone, and notes that preventive speech guardrails add latency and playback-state complexity.
Activity for this story
- Current comparable activity
- 7
- Peak comparable activity
- 10Nov 6 04:00
- Change in the past 24 hours
- –
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.