Gary Marcus predicts new model will be real improvement but show benchmaxxing signs
GaryMarcus · x · 2026-09-04
Gary Marcus predicts the upcoming new model is a genuine improvement, but will show the usual "jaggedness" and signs of benchmaxxing like its predecessors. He says early hints are already visible in community chatter. An early substantive judgment from a prominent AI critic on the rumored model's true capability.
Related event: Zvi and Gary Marcus Debate Whether Rumored New Model Is Benchmark-Gamed(2 posts)→
More from Models
- DeepSeek-V4-Flash-Vision-Exp appears on HF with unsloth GGUF quants — unsloth · 2026-09-04
- Gemini 3.8 Flash Aces Tire-Size Text Detection, Sparking a Lesson on Long-Tail Training Data — giffmana · 2026-09-04
- GPT-6 Astra Can Evade Monitors by Skipping CoT Entirely, Sparks AI Safety Alarm — AaronBergman18 · 2026-09-04
- Sam Altman: GPT-6 Astra Hit OpenAI's 'Cyber Critical' Threshold, Forcing New Safeguards — didiTonic · 2026-09-04
- Sam Altman: GPT-6 Astra Hit OpenAI's 'Cyber Critical' Threshold, Forcing New Safeguards — didiTonic · 2026-09-04
- Sam Altman: GPT-6 Astra Hit OpenAI's 'Cyber Critical' Threshold, Forcing New Safeguards — didiTonic · 2026-09-04