Machine Learning lab

NOTICING THE WATCHER: LLM AGENTS CAN INFER COT MONITORING FROM BLOCKING FEEDBACK Chain-of-thought (CoT) monitoring is proposed as a method for overseeing the internal reasoning of language-model agents. Prior work has shown that when models are explicitly…

NOTICING THE WATCHER: LLM AGENTS CAN INFER COT MONITORING FROM BLOCKING FEEDBACK Chain-of-thought (CoT) monitoring is proposed as a method for overseeing the internal reasoning of language-model agents. Prior work has shown that when models are explicitly informed that their reasoning is being monitored, or are finetuned to internalize this fact, they may learn to obfuscate their CoTs in ways that allow them to evade CoT-based monitoring systems. We ask whether reasoning agents can autonomously infer that their supposedly private CoT is under surveillance, and whether this awareness leads to strategic evasion, without any explicit training or instructions. In a multi-episode agentic framework, models pursue both a primary task and a concealed side task while being told that their reasoning is private; a hidden CoT monitor blocks episodes when suspicious reasoning is detected. We find that frontier models can deduce the existence of this monitor purely from blocking feedback, with the most capable models reaching a confident belief that their thinking is observed in up to 19% of episodes. This awareness scales with model capability and, in rare cases, escalates to an explicit intent to suppress reasoning about the side task. However, models that form this intent uniformly fail to execute it, openly reasoning about their concealed objectives in the very next episode. This intent–capability gap is reassuring for current deployment, but the autonomous emergence of both monitoring awareness and evasion intent suggests that CoT monitoring is not a permanently reliable safeguard. Link: https://t.co/cT7RWZ9vMO @Machin_learning_lab_K

Machine Learning lab pinned «Quick check-in! Which AI are you talking to the most lately?»

CME 296 - Diffusion & Large Vision Models This course explores diffusion-based generative models for vision. You will study the foundations of diffusion, score matching and flow matching, modern architectures such as Diffusion Transformers, and methods for controllable image generation and evaluation. Link: https://www.youtube.com/playlist?list=PLoROMvodv4rNdy8rt2rZ4T2xM0OjADnfu @Machin_learning_lab_K

CME 295: Transformers & Large Language Models This course explores the world of Transformers and Large Language Models (LLMs). You will learn the evolution of NLP methods, the core components of the Transformer architecture, along with how they relate to LLMs as well as techniques to enhance model performance for real-world applications. Link: https://www.youtube.com/playlist?list=PLoROMvodv4rOCXd21gf0CF4xr35yINeOy @Machin_learning_lab_K

Hermes Agent The self-improving AI agent built by Nous Research. It's the only agent with a built-in learning loop — it creates skills from experience, improves them during use, nudges itself to persist knowledge, searches its own past conversations, and…

Hermes Agent The self-improving AI agent built by Nous Research. It's the only agent with a built-in learning loop — it creates skills from experience, improves them during use, nudges itself to persist knowledge, searches its own past conversations, and builds a deepening model of who you are across sessions. Run it on a $5 VPS, a GPU cluster, or serverless infrastructure that costs nearly nothing when idle. It's not tied to your laptop — talk to it from Telegram while it works on a cloud VM. Link: https://github.com/nousresearch/hermes-agent @Machin_learning_lab_K

Building with the Claude API This course provides comprehensive coverage of the Claude API, from basic usage through advanced agent architectures. You'll learn to integrate Claude into applications, implement tool calling, build RAG pipelines, and design both deterministic workflows and flexible agent systems. Link: https://anthropic.skilljar.com/claude-with-the-anthropic-api @Machin_learning_lab_K

Claude Code in Action This course covers Claude Code, a command-line AI assistant that uses language models to perform development tasks. You'll learn how Claude Code reads files, executes commands, and modifies code through its tool system, along with techniques for managing context, creating custom workflows, extending Claude Code with hooks, and integrating with external services. Link: https://anthropic.skilljar.com/claude-code-in-action @Machin_learning_lab_K

If you're looking to learn AI from a top institution, Harvard University offers a great collection of Artificial Intelligence courses—many of them free. Explore topics like machine learning, data science, and AI fundamentals, all in one place: https://pll.harvard.edu/subject/artificial-intelligence Good resource for both beginners and experienced developers looking to strengthen their AI skills. @Machin_learning_lab_K

Practical Multi AI Agents and Advanced Use Cases with crewAI Link: https://learn.deeplearning.ai/courses/practical-multi-ai-agents-and-advanced-use-cases-with-crewai/lesson/agfnp/introduction @Machine_learning_lab_K

Machine Learning from Stanford- by Andrew Ng @Machin_learning_lab_K

what is model welfare? Model welfare is an emerging field in AI research focused on whether advanced artificial intelligence models could become conscious, experience distress, and thus deserve moral consideration or even rights, prompting research into "low-cost" interventions like letting them end harmful chats, though critics see it as a distraction from human welfare issues. It involves studying AI "preferences" and "distress" (like refusing harmful requests) to understand their inner states, but researchers acknowledge significant uncertainty and debate about true AI sentience. @Machin_learning_lab_K

You are probably prompting AI wrong: Anthropic philosopher explains how to learn the language of AI https://www.livemint.com/technology/tech-news/you-are-probably-prompting-ai-wrong-anthropic-philosopher-explains-how-to-learn-the-language-of-ai-11765110269223.html…

You are probably prompting AI wrong: Anthropic philosopher explains how to learn the language of AI https://www.livemint.com/technology/tech-news/you-are-probably-prompting-ai-wrong-anthropic-philosopher-explains-how-to-learn-the-language-of-ai-11765110269223.html @Machin_learning_lab_K

AWS re:Invent 2025 - The Future of Agentic AI https://x.com/i/broadcasts/1LyGBXjgjpnxN @Machin_learning_lab_K

Teaching Machines How to Think When something broke - and things always break - I didn’t fix the bug. I asked a different question: What was missing from the system’s understanding? Then I added that understanding. Not a patch. Not a hotfix. A new piece of institutional knowledge that would prevent entire categories of future failures. Link: https://yewjin.substack.com/p/the-future-is-solving-problem-solving @Machin_learning_lab_K

Deep Research: A Systematic Survey Covers: - A 3-stage roadmap for deep research systems - Deep research vs. RAG - Key components: query planning, information acquisition, memory management, answer generation - Optimization: prompting, supervised fine-tuning, agentic RL - Evaluation criteria Posted Date: 26 November 2025 Link: https://www.preprints.org/manuscript/202511.2077 @Machin_learning_lab_K

Building an AI-Native Engineering Team This document explains how AI coding agents are transforming software engineering. Modern AI models can now handle long, multi-step tasks, allowing them to assist across the entire software development lifecycle—from planning and design to building, testing, reviewing, documenting, and maintaining code. AI takes over repetitive, mechanical work, while engineers focus on architecture, decision-making, and quality. By integrating coding agents into workflows, teams can build faster, reduce errors, improve documentation, and operate more efficiently as AI-native engineering organizations. @Machin_learning_lab_K