I sat down with AI expert Gary Marcus, Professor Emeritus at New York University, at Black Hat 2025, where he broke down the release of GPT-5 and how it compares to GPT-4. He explains why the upgrade is incremental rather than revolutionary, why it’s still plagued by hallucinations and reasoning errors, and why it remains far from achieving Artificial General Intelligence. He also points out that GPT-5 performed worse than Elon Musk’s Grok 4 on the ARC-AGI-2 benchmark and discusses what this means for the AI race.

#chatgpt5 #ai #bhusa2025

David Bombal

AI expert Gary Marcus expresses disappointment with GPT-5's release, arguing it fails to meet the transformative expectations set over 32 months of anticipation. He notes that while GPT-5 shows improvements, it remains part of the competitive pack rather than a breakthrough, with Elon Musk's Grok 4 outperforming it on certain benchmarks. Marcus emphasizes that GPT-5 still suffers from fundamental issues like hallucinations and reasoning errors, falling short of the artificial general intelligence (AGI) that many hoped for.

AI Expert Gary Marcus on the Harsh Truth About ChatGPT-5