Deep Research dissected: slow, expensive, beautiful — and dangerously convincing
TobyWalsh · x · 2026-08-29
AI researcher Toby Walsh shared this long read on Deep Research, wryly noting it would have been much better had AI not been heavily used to write it.
Key points of the article:
- Deep Research reports are flawlessly structured and heavily cited — a 30-page document with 80 footnotes looking like a week of analyst work, which is exactly what makes them both beautiful and dangerous.
- It cites a Reddit user's case: a polished, well-cited report on European regulation whose sources fell apart when spot-checked before a meeting — looking true is not being true.
- The author calls Deep Research a flattery generator trawling the internet for garbage, examines why it is slow, expensive, beautiful and dangerous, and warns paying users not to relax verification just because the output is polished.
More from Models
- GLM-5.3 Flash costs 17x less with slight accuracy drop — zainhas · 2026-08-29
- Moore Threads MTT S5000 achieves Day-0 support for Zhipu GLM-5.3-Flash — teortaxesTex · 2026-08-29
- Nous Research Offers Free Models on Nous Portal Platform — Teknium · 2026-08-29
- Open source eval controversy: lab accused of creating false hope — teortaxesTex · 2026-08-29
- Tencent Hy4 Tops SWE-bench Pro, Now Available in Cline — TencentHunyuan · 2026-08-29
- User calls out Gemini 3.7 for compulsive lying and poor work ethic — AlexandraMaryWindsor · 2026-08-29