Google NotebookLM Evolves Into an Agentic Research Tool Powered by Gemini 3.5 Flash
Google has significantly upgraded its research tool NotebookLM. The new version is built on Gemini 3.5 Flash, can run code on a dedicated cloud computer, and includes agentic capabilities that autonomously discover sources through Google Search. In internal testing, the new system reportedly outperformed the previous version in up to 78.2% of cases.
NotebookLM has primarily been used for summarizing sources you feed it, so its evolution into an agentic tool capable of autonomously searching, analyzing, and even executing code is noteworthy. It has the potential to significantly change workflows for researchers and students.
Ads Begin Rolling Out to ChatGPT's Free and Go Plans
On June 10, OpenAI revised ChatGPT's privacy policy, adding provisions covering advertising on the free plan and the "Go" plan. The change affects free users and Go plan subscribers, while paid Plus and higher plans are expected to remain unaffected. The policy update takes effect on June 22.
ChatGPT's free user base is said to number in the hundreds of millions, making ad revenue a significant monetization channel for OpenAI. However, given concerns about user experience and privacy, how ads are displayed and how frequently will likely draw close scrutiny.
Germany Establishes AI Safety Institute "DE-AISI"
Germany's National Security Council has approved the establishment of an AI safety institute, DE-AISI, modeled on the UK's AISI (AI Safety Institute). DE-AISI plans to test frontier models from companies such as Anthropic and OpenAI for security risks.
At the same time, critics point to a fundamental problem: with no homegrown frontier models within the EU, dependence on US and Chinese AI technology will persist. With model providers closely tied to their own governments, the question of how Europe can secure its digital sovereignty remains pressing.
China's MooreThreads Releases Open-Source Coding Model "MusaCoder-27B"
Chinese GPU maker MooreThreads has released MusaCoder-27B, an open-source coding-focused model, on Hugging Face. Details of the model are available in an arXiv paper.
The number of models optimized for inference on Chinese-made GPUs is growing, and an ecosystem built on vertical integration of hardware and software is taking shape. It is a trend worth following as China's AI industry seeks to break away from its dependence on NVIDIA.
Notable Discussion: Is Symbolic AI Becoming Important Again?
Heise recently featured a discussion titled "Is symbolic AI more important than ever?" which has also gained traction on Hacker News. With LLM hallucinations and reasoning reliability remaining key challenges, interest in hybrid approaches that combine neural networks with symbolic reasoning is on the rise.