OpenAI and MIT say GPT-5.6 agent autonomously calibrates superconducting qubits
新智元 · wechat · 2026-09-09
A joint OpenAI–MIT (EQuS) case study reports that GPT-5.6 running on a Codex agent framework performed end-to-end autonomous quantum computing experiments: controlling lab hardware, emitting microwave pulses, cleaning and fitting data, and iteratively calibrating a never-measured 6-qubit superconducting chip with only 4 human interventions across 40 routine measurements. Researchers supplied experimental context and measurement 'skills'; limits remain at low signal-to-noise and anomalous physics, where expert humans still outperform. The piece also mentions OpenAI's claimed Navier–Stokes result (10,000 agents, 88 hours, 130B tokens), disputed by mathematician Tristan Buckmaster as rushed and actually solving Euler singularities.
More from Research
- Innocent-looking chain-of-thought can hide AI misbehavior from monitors, research finds — evilsocket · 2026-09-09
- Cross-cultural curiosity paper on human vs LLM questioning accepted at AACL 2026 — radamihalcea · 2026-09-09
- Michigan PhD student Jae-Won Chung makes MIT TR35 for Zeus, an AI energy-efficiency tool — radamihalcea · 2026-09-09
- Thread follow-up: the tail-area criterion traces back to 1990s Bayesian literature — BlackHC · 2026-09-09
- SGD doesn't inherit the Bayesian identity — summed losses are a prequential score — BlackHC · 2026-09-09
- Why loss curves cross: misspecification, complexity, and prior fit — BlackHC · 2026-09-09