GPT Pro user finds model browses completely irrelevant sources, questions real retrieval
CompetentRaindeer · reddit · 2026-09-20
A $200-tier Pro subscriber asked the model to gather LLM image-prompt best practices, then inspected its browsing trail: every cited link was unrelated to the task, with clickbait-y titles and off-topic content. The user questions whether the model actually reads these sources or is faking retrieval. The post also plugs his Windows desktop tool Kiln for managing skills/configs across Codex, Claude, GH Copilot, plus turning YouTube videos into testable prompt skills.
More from Models
- Gemini Flash 3.8 Goes Off the Rails: Prescription Scan Triggers Random Math — MaziyarPanahi · 2026-09-20
- LLM vs classical ML across 8 datasets: labeled data still favors SVM and XGBoost — Ok_Juggernaut2187 · 2026-09-20
- Researchers Use Claude Opus 5 to Hijack OpenAI Employee Accounts for Under $3,000 — FinanceYF5 · 2026-09-20
- NetEase Youdao open-sources Confucius4-R2T2 streaming ASR with 200-600ms latency — aigclink · 2026-09-20
- Alibaba open-sources DAMO RADAR medical model detecting cancer and ~150 conditions — emmanuelvivier · 2026-09-20
- Anthropic reportedly testing Claude Money to link bank accounts for spending analysis — emmanuelvivier · 2026-09-20