TechFlow Logo
Login/ Sign up
ETH Gas
Gwei
Fear
gas
7x24hNews

OpenAI Chief Scientist Responds to Astra Controversy: Chain-of-Thought Monitoring Remains a Research Focus

2026.09.02 - 06:20
Share

TechFlow reports that on September 2, OpenAI Chief Scientist Jakub Pachocki expressed hope to avoid a race toward making models "unmonitorable" due to misleading reports. He stated that current frontier models, including Astra, have a computational graph depth no more than twice that of GPT-4. Since the release of its first reasoning models, OpenAI has been committed to retaining and leveraging chain-of-thought monitoring, believing that this technology helps observe how model alignment capabilities generalize from the training distribution.

Pachocki also pointed out that while chain-of-thought monitoring is relatively fragile and the trend is not optimistic, it can still be strengthened through research and has become a core objective of OpenAI's current research program.

Previously, The Information reported that Astra uses recurrent depth, allowing the same set of Transformer layers to compute repeatedly. Part of the reasoning can therefore occur internally within the model without needing to be fully written out as a chain of thought, which also sparked concerns that models may become increasingly difficult to monitor.

Add to Favorites
Share to Social Media