Researcher slams LLM-generated BART vs Tesla energy chart: apples-to-oranges data
JFPuget · x · 2026-08-28
Jean-François Puget publicly criticized a BART-vs-Tesla energy comparison chart (with data apparently pulled by LLMs), arguing it makes a classic data science error: BART's consumption includes offices, station lighting, and track maintenance, while the car side only counts driving energy — not an apples-to-apples comparison. He also questioned whether Tesla figures used highway cruising rather than city driving, and noted that Gemini and GPT 5.6 Sol returned very different numbers, a red flag for jumping to conclusions. Grant Slatton countered that transit agencies publish passenger-miles and electricity use, giving 362 Wh/passenger-mile. The debate itself is a live case study in LLM data-analysis reliability.
More from Fun
- Dev Shares Vibe Coding Stack: Codex, Claude Code, Grok and OpenCode — Al_Grigor · 2026-08-28
- If we misuse the 'AI is like nuclear technology' meme, LinkedIn is definitely Chernobyl — curious_vii · 2026-08-28
- Generating Jellyfish in p5.js with Claude Opus 5 — anselm · 2026-08-28
- Simon Willison's LLM cliché highlighter now covers 38 patterns — sebpaquet · 2026-08-28
- The 'cataphoric teaser': the linguistic tell of AI text creating cheap suspense — sebpaquet · 2026-08-28
- Turn Your GitHub Profile Into a GTA VI Style City — filiksyos · 2026-08-28