AI & Materials
Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.Cases where AI was actually used in materials research and scientific experiments.
매일 00·06·12·18시(한국시간)에 웹에서 새 소식을 찾습니다. AI 활용 방법과 결과가 확인된 새 사례만 게시합니다.
Checking notifications…
A npj Computational Materials paper used a diffusion framework with a shared score network to jointly generate crystal structures, Bader charge, and atomic DOS for inverse design. It reported improved success rates and physical plausibility across multiple conditions compared with a structure-only baseline.
SLAC was selected as a project in the DOE Genesis Mission and will advance a closed-loop autonomous catalyst discovery workflow that combines real-time X-ray measurements with evidence and hypothesis ranking. The goal is to implement and validate a research system that accelerates catalyst understanding and design.
A Nature Computational Science paper addressed LLM-based property prediction using crystal structures and categorical information as inputs, and presented design guidance that a smaller model may work better than a larger model. This is closer to a practical guide for model selection and design than to new materials discovery.
Google DeepMind said on 2026년 10월 13일 that it has begun providing Gemini 4 Argon first to real defenders before expanding distribution more broadly. The announcement focused on cyber defense, prompt injection defense, safety evaluation, and safeguards, revealing a stepwise approach to model rollout management.
Mistral announced the new large model Mistral Large 4 on 2026년 10월 6일. The company said the model is trained and served on its own European infrastructure, with a focus on security, cyber work, and enterprise use. Competition among large models is expanding beyond performance to include operational sovereignty and deployment methods.
A study published in npj Computational Materials on 2026-10-15 performed inverse design by a method that jointly generates crystal structures and local electronic descriptors. While detailed quantitative results are limited in the public information, the workflow confirms the generation of material candidates to find target properties.
An OpenAI newsroom announcement dated 2026-10-07 공개ed the product direction related to GPT-6 and Intelligent UI. Around the same time, 2026년 10월 updates for GPT-6 Sol and GPT-6 Luna were also posted, confirming the expansion of the line-up together.
A Scientific Reports study compared ridge regression, random forest, extra trees, and gradient boosting to predict the optical bandgap and photocatalytic performance of rare-earth-substituted spinel ferrites. Among the candidate models, random forest showed the lowest error, but MAE 21.43 percentage points and R² = -0.56 indicated limited generalization performance.
The PRL paper proposed a thermodynamically consistent machine-learning potential (TCCP-MLP) for constant-potential interactions. Trained only on energy and forces, it computes the work function as a derivative of a consistent energy function and was used to predict the voltage-dependent activation barrier and charge transfer for C–C bonding on the Cu(100) surface.
Anthropic said that as of 2026-10-15 it has begun granting employee-like access to the nonprofit evaluation organization METR. The move is intended to more strongly verify frontier models and safety practices, alongside evaluation work with SecureBio and Accenture and additional discussions with Hugging Face and others.
A self-driving lab incorporating machine learning was combined with microfluidic experiments to rapidly optimize the synthesis conditions of multicomponent double perovskite nanoplatelets. It showed that data-driven autonomous exploration is possible even in complex compositional systems containing up to 6 elements.
An autonomous closed-loop framework combined active learning and quantum modeling to search for passivation molecules for perovskite solar cells. With 5ANI, high efficiency was reported in both small-area cells and mini-modules, together with long-term stability.
Anthropic said on 2026-10-14 that it opened a new office in Singapore and signed an MOU with the Singapore government for collaboration on AI safety and evaluation. Around the same time, the item appeared as the latest announcement in the official newsroom, indicating that regional expansion and policy and evaluation collaboration are being advanced together.
OpenAI announced on 2026-10-07 that GPT-6 was being deployed to a broader user base and that Intelligent UI had been introduced. This feature immediately generates an interactive UI together with the answer, helping learning, exploration, and task execution in a visual way. In the same announcement, OpenAI said GPT-6 has stronger jailbreak resistance and defenses against high-risk misuse.
Researchers rapidly evaluated 106,235 Heusler structures in DXMag using an atomistic line graph neural network that predicts spin-resolved d-DOS, and narrowed the thermoelectric candidates together with first-principles transport calculations. The predicted DOS was validated by comparison with DFT DOS, and 80 computational candidates were confirmed.