2 min read
In short
Today’s roundup includes a data storage issue in Google AI Studio, a new test of autonomous agents for industrial automation, and useful community projects.
Today’s roundup includes a data storage issue in Google AI Studio, a new test of autonomous agents for industrial automation, and useful projects from the community.
🔥 Hot:
🔹 Data persistence after deletion was discovered in Google AI Studio — The issue is marked as critical in the tracker; there are still few details about the incident’s scope. 🔹 PLCBench tests whether LLM agents can turn PLC access into prolonged physical actions — The study focuses on the risks of autonomous agents working with industrial controllers.
➡️ News:
🔹 Claypot launched a block-based builder for teaching children about AI systems — The project lets users assemble models from blocks and observe how they work. 🔹 ScriptTap automates Android without root access through AI scripts — The app uses a script contract to control actions on the device. 🔹 Veryfront Code builds full-stack AI applications and agents with TypeScript and React — The project provides a ready-made foundation for developing applications with agents. 🔹 Framework added a configuration with 192 GB of memory — This amount is particularly interesting for running large language models locally.
➡️ Useful resources:
🔹 An sm_120 kernel was built for FlashMLA with a claimed 2–3× speedup — The author adapted the kernel for a new GPU architecture to speed up training and inference for models with MLA. 🔹 Internet Archive compiled a Vintage AI collection — The archive offers historical materials on the development of artificial intelligence.
➡️ Discussions and case studies:
🔹 A user proposed solving the problem of separate AI tokens across services — Currently, the limits of one product cannot be used in another: tokens are tied to separate subscriptions and tools. 🔹 The community compared Qwen 3.8 Flash Next and GLM 5.3 Flash for image generation — An informal test showed differences in how accurately the models reproduce the scale and details of the source image.
📝 If you’d like to add other news and resources to the list, write in the comments.