#chain of thought — The Focus Digital

SEC.03 / LLM Research [01] 2026.09.06

A dim lab bench with a laptop showing a sparse reasoning trace next to a printed safety checklist

GPT-6 Astra Hits Critical Cyber. Its Thoughts Got Harder to Read

OpenAI's Astra is the first Critical-cyber model they broadly deployed. Recurrent-depth loops cut readable chain of thought. BenchMIRT shows why scores mislead.

#LLM#OpenAI

READ →

SEC.03 / LLM Research [02] 2026.07.10

An editorial illustration showing an AI neural network tangled in its own thought chains, with glowing reasoning paths that loop back on themselves and a simple correct answer sitting just out of reach, rendered in deep indigo and warm amber tones

LLMs Can Think Too Much — And It's Making Them Wrong

The AI field spent two years proving that longer reasoning chains produce better answers. Now a wave of new research shows the opposite: beyond a certain point, more 'thinking' actively degrades performance. The overthinking problem is real, measurable, and it's forcing a rethink of how we deploy reasoning models.

#LLM reasoning#test-time compute

READ →

SEC.03 / LLM Research [03] 2026.06.24

The Inference-Time Revolution: How Test-Time Compute Scaling Is Rewriting the Rules of LLM Intelligence

Pre-training scale isn't the only lever anymore. Test-time compute scaling lets models think longer at inference — and the results are reshaping how we measure AI capability.

#test-time compute#LLM reasoning

READ →