No benchmark degradation isn't evidence of absence, argues AI researcher

maier_ak · x · 2026-10-01

maierak adds to the planned-obsolescence discussion: the hidden knob sits upstream of every benchmark outsiders can run, so "unchanged" and "turned-down" produce identical tables. Absence of degradation on public benchmarks is not evidence of absence — Volkswagen passed every official test too.

Related event: Researchers Warn AI Vendors Could Silently Degrade Models, Like the Phoebus Cartel(3 posts)→

Original post →

More from Models

• Developers claim unreleased Gemini 4 Argon scores 91.9 atop VibeCodeBench — sunjiao123sun_ · 2026-10-01
• Claude Opus 5.5 paints Qi Baishi-style ink shrimp in live WebGL shaders — anselm · 2026-10-01
• Did Opus 5.5 enter a "nerfed" phase? LiveNerf baseline update fuels debate — Kazaan · 2026-10-01
• OpenAI halves Pro 200 included usage at same $200 price, launches $500 Pro 500 tier — Mr_LA · 2026-10-01
• Hands-on: binary quantization is lossy for most models, ternary works fine — antoine_chaffin · 2026-10-01
• RLHF co-author Diogo Almeida discusses the philosophy behind TypeSafe's Jev model — AxSaucedo · 2026-10-01

Models channel →