Agent tested on real trip-planning from email: ChatGPT nails trains, gives up on hotel availability
3scorciav · x · 2026-09-16
A user benchmarked big-model agents on the canonical task "plan my trip from today's email" with all connectors enabled. ChatGPT (Work) browsed well and found train connections and good hotels, but failed to check room availability and pricing at two hotels before giving up. The author also speculates Gemini's experience may differ vastly between free, paid subscription, and API tiers.
Related event: Hands-on test: AI travel agents stumble on hotel booking(3 posts)→
More from Fun
- Wikihack Hits HN Front Page: HackerNews-Mentioned Wikipedia Articles, Curated — Aizkmusic · 2026-11-02
- xhdr.app: free local tool turns any photo into an HDR profile picture for X — jaivin · 2026-11-02
- Creator uses Codex to build a particle body in code, pulsing to a Suno-generated track — OpenAIDevs · 2026-11-02
- AI-generated 'Toblerone traffic cone' and other absurdly real product concepts — Delahuntagram · 2026-11-02
- Microsoft engineer marks her 'half-life': half her life now spent at Microsoft — jennifermarsman · 2026-11-02
- AI product concept: an R2-D2-shaped kettle that looks production-ready — Delahuntagram · 2026-11-02