Gradient Dissenter
Works at METR and evaluated GPT-5 before its external deployment. Cross-posted the blog post about METR's evaluation of GPT-5 to LessWrong.
Best podcasts with Gradient Dissenter
Ranked by the Snipd community
4 snips
Aug 8, 2025 • 48min
“METR’s Evaluation of GPT-5” by GradientDissenter
Gradient Dissenter, who works at METR and played a key role in evaluating GPT-5, discusses the thorough safety analysis conducted on the AI model prior to its launch. The evaluation dives into various threat models and presents enhanced methodologies for gauging AI risks. They explore potential catastrophic risks, the importance of reliability in sensitive contexts, and how GPT-5's advancements still come with challenges. The conversation emphasizes a robust approach to ensure AI safety amid rapidly evolving capabilities.
