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…
Anthropic said it has started granting employee-level access to the nonprofit evaluator METR to better assess the safety of its systems. The company also said it is in early discussions with other organizations, including Hugging Face. It reads as a signal that external independent evaluation is being integrated more deeply into actual operating procedures.
A Physical Review Applied paper reported using SVD-based dimension reduction to shrink a high-dimensional pulse parameter space and perform closed-loop quantum gate calibration with noisy experimental data. It is a case of designing and calibrating an Xπ/2 gate that is robust to errors and spectator qubit disturbance in a high-dimensional control space.
This study reconstructed liquid-liquid phase diagrams and applied them to extraction process design using a physics-informed learning framework. It can be viewed as a case in which physical constraints were built into the learning process to address both phase-equilibrium problems and process design.
Researchers used physically interpretable descriptors to design and screen candidate diluents for lithium metal battery electrolytes with machine learning. This is an example of combining interpretable features and a prediction model in battery electrolyte composition search.
Anthropic has expanded its cyber verification program and launched a new framework that gives verified security experts higher cyber capabilities and more permissive blocking classifiers. Security teams can now apply for access tiers aligned with their scope of work, and the previously fragmented access procedures have been unified into a single system.
Google announced that Gemini 4 Argon will be made available to developers on 2026년 10월 13일. The model is available in Google AI Studio, Android Studio, the paid Gemini API, and Google Antigravity, and is introduced as a frontier model focused on complex real-world workflows and multi-step tasks.
Physical Review B 논문은 텐서 네트워크로 만든 전하 안정성 데이터에서 시각 기반 신경망을 사용해 반도체 양자점 2차원 스핀 큐비트 배열의 무작위 onsite-energy fluctuation을 추정했다. 3×3과 5×5 배열에서 높은 R²를 보고했으며, 중심 영역 관측만으로 핵심 파라미터 추정 가능성을 제시했다. 다만 시뮬레이션 기반 이론 연구다.
OpenAI announced GPT-6.1 Sol and began offering it immediately to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work, as well as to API users. The model was highlighted for improving on GPT-6 Sol and for delivering results close to Astra on some task types at a lower cost.
MIT presented a multimodal vision-language model that jointly learns 3D atomic structures and paper text, and reported that it found better crystalline materials candidates than a text-only model. In a solid electrolyte query, independently known promising candidates were included among the top results, and structure-text alignment was also improved.
Anthropic revised its usage policy on October 8, 2026 to clarify controls on deepfakes, election disruption, and high-risk use, and added further requirements for hardware-coupled situations in which physical actions are performed autonomously.
OpenAI said it released the October update for GPT-6 Sol and GPT-6 Luna on October 7, 2026, and carried out a full deployment of ChatGPT. The update strengthened defenses against high-risk misuse related to cyber, biological, and violence, and the company said jailbreak resistance, deception, and evasiveness were also reduced.
Anthropic announced that it is expanding asset allocation for public benefit. The company said the share of equity allocated to the public benefit will increase further and may grow more depending on additional valuation conditions.
Anthropic said it has begun providing employee-like access to the independent evaluator METR. The move is intended to allow deeper external assessment of frontier model risks, and METR will publish the evaluation results.
Google DeepMind announced that it will develop a fine-tuned AI model with the Singapore government to detect fraud in messaging environments. The goal is to strengthen scam prevention without infringing on personal privacy.
Google DeepMind unveiled Gemini 4 Argon on 2026년 10월 13일. It said the latest frontier model is available to developers, enterprises, researchers, and consumers, and can be used immediately in API and multiple Google products. It also presented performance and pricing information for complex tasks spanning code, finance, law, cybersecurity, and science.