![]() Anthropic warns it is very close to RSI capability, aka self-building AIThe company stated in its S-1A filing that the acceleration of progress driven by artificial intelligence developing its own models carries security risks. Anthropic emphasized that no model has yet crossed this threshold, but it plans to maintain human oversight. |
AiPulse — Issue No. 107
COVER
At Work
Two researchers were awarded for discovering the first AI-discovered antibiotic, halicin, using deep learning and for their pioneering work in this field. This achievement is seen as a milestone demonstrating the potential of artificial intelligence in developing new drugs effective against resistant bacteria. |
The article argues that aging is not only a medical but also an architectural issue. It proposes designing a built environment that goes beyond the architectural standards’ assumption of an “ideal user,” one that does not restrict the independence of the elderly or make them dependent on institutions. |
MPC and Reinforcement Learning Compared in Renewable Hydrogen ChainThe study simulated model predictive control and reinforcement learning methods within physical constraints to manage the hydrogen supply chain under uncertainty. The results showed that model predictive control provided the highest economic efficiency when reliable forecasts were available. |
Global Lens
The authority implemented interim measures amid suspicions that Google's 'AI Overviews' and 'AI Mode' features were excluding publishers. The investigation will examine how these features affect content flow and user access. |
The judicial administration in South Korea is seeking a legal basis to address fake court rulings fabricated with artificial intelligence in the petitions filed by lawyers and parties. The institution will submit its opinion to the National Assembly regarding legal regulations that propose imposing fines to prevent this situation. |
SenseTime released the multimodal AI model SenseNova Loop-9B-Preview as open source under the Apache 2.0 license. The model increases processing depth by running the 24-layer block twice without copying the weights. |
The Abu Dhabi administration has launched an AI-supported pilot application for food import compliance inspections using OpenAI models. This system aims to accelerate the decision-making processes of public officials and to make regulations machine-readable. |
Apple will integrate Alibaba's Qwen model into Siri and text tools in the iOS developer beta starting November 4. These features will be rolled out to a broader audience in December with iOS 27.3. |
Anthropic's artificial intelligence model Claude demonstrated the potential to accelerate the development of antitoxins by generating miniproteins that can bind to neurotoxins in scorpion venom. This study proves that artificial intelligence can offer faster and lower-cost solutions in biomedical research compared to traditional methods. |
Fun & Curious
A Visier study revealed that 48 percent of employees overstate their use of artificial intelligence. Amazon acknowledged that these metrics were disconnected from the essence of the job and retracted its measurement approach. |
OpenAI developed Pando, a custom Git backend, to manage the massive code traffic generated by engineers and agents. This system eliminated infrastructure bottlenecks by reducing average latency by 80% and increasing reliability to 99.99%. |
FOLLOW
Get new issues in your feed reader — add this to an app such as Feedly, Inoreader or NetNewsWire. Follow by RSS |










