An honest LLM experiment: using web-search news analysis to time dollar exchanges didn't work
zit-hb · reddit · 2026-11-10
The author received some US dollars and wondered whether an LLM constantly analyzing the news could help pick a good time to exchange them. The result: it couldn't. Still, they believe the approach has real potential for compressing public perception into an easy-to-read value, and they've adjusted the setup and upgraded to GPT-5.6 to keep testing. An honest negative-result writeup useful for anyone exploring LLM-driven news sentiment quantification.
Related event: Test Finds LLM News Analysis Fails to Predict Dollar's Exchange Rate(2 posts)→
More from Research
- Two New Surveys Land: Continuous Diffusion LMs and Mechanistic Interpretability for AI Safety — burny_tech · 2026-11-11
- Meta ships SAM 3D Body DINOv3 for 3D human mesh recovery on Hugging Face — facebook · 2026-11-11
- AI2 open-sources ToyBlocks to benchmark robotic world models by task-relevant fidelity — allen_ai · 2026-11-11
- Zero-dependency MCP hub fronts nine research repos as 14 tools without merging dependency trees — Emergency-Tutor-7 · 2026-11-11
- OpenTac: open-source ecosystem turns any robot surface into a tactile skin — _krishna_murthy · 2026-11-11
- tts-bench round 3: 91 open TTS models and 28 paid APIs benchmarked on identical prompts — UkieTechie · 2026-11-11