Zvi and Gary Marcus Debate Whether Rumored New Model Is Benchmark-Gamed

Discussing a rumored new model, Zvi argued the AI community would expose benchmark-gaming within days, while Gary Marcus predicted the model will show real improvement but again exhibit jagged, benchmark-focused weaknesses.

2026-09-04 ~ 2026-09-04 · 2 related posts

• Zvi: If the rumored model is benchmaxxed, the AI community will spot it within days — TheZvi · 2026-09-04
• Gary Marcus predicts new model will be real improvement but show benchmaxxing signs — GaryMarcus · 2026-09-04