Are ML Papers Getting Too Wordy? Reddit Debates 40-Page Bloat
NeighborhoodFatCat · reddit · 2026-10-11
A Reddit thread argues ML research has become "wordy": papers routinely run 20-40+ pages of text, even outside LLM topics. The author cites three issues: bloated length (e.g., the LLaMA paper), undefined or pseudo-new terminology with conversational tone creeping into publications (naming Anthropic's transformer-circuits framework post), and a decline of rigorous math in favor of block diagrams that are nearly impossible to translate into math or code without guesswork. The thread sparked debate on academic writing norms and review standards in the LLM era.
More from Research
- 1,680 A/B trials show 50-year-old Unix formalisms crush English prompt rules, 98.3% vs 0.8% — RandalSchwartz · 2026-10-11
- Benchmark audits find ~30% of SWE-Bench Pro tasks broken; SciCode and Terminal-Bench flawed too — geoffwolfe · 2026-10-11
- 17 researchers led by ex-Huawei Noah's Ark head publish definitive guide to building AI agents — blaizedsouza · 2026-10-11
- 3 papers on adversarial AI attacks vs flight control, CCTV, 6G at IEEE conf — WeisiG · 2026-10-11
- Anders Sandberg's Claude-made video shines with Tromp's Lambda Diagrams notation — anderssandberg · 2026-10-11
- Lean Co-creator Jeremy Avigad on AI, Verification, and the Future of Mathematics — EchoShao8899 · 2026-10-11