I have just came across this info in my LinkedIn feed posted by an AI entrepreneur from Germany (putting here the translation from German) May be helpful if you have not updated your versions on n8n: “Critical security vulnerability in n8n! 17,546 German servers affected! n8n is a workflow automation platform. A vulnerability currently exists that could give attackers complete control over the server. The problem: n8n executes code in workflows without adequately isolating it. An authenticated user can therefore execute arbitrary commands on the server. Possible consequences: Full system access, stolen credentials, access to databases and APIs. The numbers: - 103,963 vulnerable servers worldwide - 17,546 in Germany (ranked 2nd) - CVSS score 9.9 (Critical) - Exploit already publicly available What to do now: Update to version 1.120.4, 1.121.1, or 1.122.0. Ideally immediately. The exploit is public. Every day without a patch is an avoidable risk.”

Worth reading and keeping an eye on.. Introducing Nested Learning: A new ML paradigm for continual learning We introduce Nested Learning, a new approach to machine learning that views models as a set of smaller, nested optimization problems, each with its own internal workflow, in order to mitigate or even completely avoid the issue of “catastrophic forgetting”, where learning new tasks sacrifices proficiency on old tasks. The last decade has seen incredible progress in machine learning (ML), primarily driven by powerful neural network architectures and the algorithms used to train them. However, despite the success of large language models (LLMs), a few fundamental challenges persist, especially around continual learning, the ability for a model to actively acquire new knowledge and skills over time without forgetting old ones. When it comes to continual learning and self-improvement, the human brain is the gold standard. It adapts through neuroplasticity — the remarkable capacity to change its structure in response to new experiences, memories, and learning. Without this ability, a person is limited to immediate context (like anterograde amnesia). We see a similar limitation in current LLMs: their knowledge is confined to either the immediate context of their input window or the static information that they learn during pre-training. The simple approach, continually updating a model's parameters with new data, often leads to “catastrophic forgetting” (CF), where learning new tasks sacrifices proficiency on old tasks. Researchers traditionally combat CF through architectural tweaks or better optimization rules. However, for too long, we have treated the model's architecture (the network structure) and the optimization algorithm (the training rule) as two separate things, which prevents us from achieving a truly unified, efficient learning system.

Amazing video!

Google just dropped 40+ page guide on taking your AI Agent from prototype to production. 100% free. 🙌🏻

I had a conversation with an Indian guy, who has built an agent that interviews IT people for jobs, and he was going at vibe coding giving examples of how it can be completely insecure and that they prefer to build things coding themselves without any AI tools. Even though I think he was taking it to far, this actually made me think about the potential security issues and how to prevent them. I would really appreciate if guys who came across this issues or prevented them shared their experience. From my side, I found some videos and sources to study this subject for myself. Sharing a talk I found: https://www.youtube.com/watch?v=XaosRsgGSX8