OpenAI o1 Reasoning Model GPQA Benchmark
2 videos · score: 2,155 · first seen Jun 22, 2026
OpenAI's new o1 Reasoning Model has sparked discussion after achieving human-level performance on the GPQA benchmark in STEM fields, with Tejal Patwardhan highlighting its paradigm-shifting potential in compute efficiency, while creators like Manuel Arias raise concerns about AI's growing role in compromising DeFi security through low-cost, high-impact exploits.
The AI Exploit That Could Destroy DeFi
Manuel Arias's May 2026 declaration that AI agents have made DeFi structurally unsafe, citing $1.22-per-contract exploit costs and the record $630 million April 2026 DeFi exploit month (including the KelpDAO bridge hack triggering an $8.45 billion Aave bank run), argues that even blue-chip protocols like Aave are now vulnerable.
Why Tejal Patwardhan stopped underestimating the models - Episode 21
Tejal Patwardhan reveals that OpenAI discovered in fall 2023 that a model trained only on math achieved PhD-level performance on the GPQA benchmark, which she frames as a proof point for the reasoning paradigm shift rather than the end goal.