Here are some recommended (not required!) reading for the course. This list contains articles that either the course organizers or the lecturers found interesting. Scientific articles The DeepSeek-R1 paper provides an interesting and surprisingly approachable overview of the development of the currently-best open weights large language model. DeepSeek-AI, Guo, D., Yang, D., Zhang, Haowei, Song, J., […], Zhang, Zhen, 2025. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning. https://doi.org/10.48550/arXiv.2501.12948 In a now-classic paper, Mnih et al. demonstrate that complex behaviors (beating video games, in this case) can arise from reinforcement learning on a relatively simple model. Mnih, V., Kavukcuoglu, K., Silver, D., Rusu, A.A., Veness, J., Bellemare, M.G., Graves, A., Riedmiller, M., Fidjeland, A.K., Ostrovski, G., Petersen, S., Beattie, C., Sadik, A., Antonoglou, I., King, H., Kumaran, D., Wierstra, D., Legg, S., Hassabis, D., 2015. Human-level control through deep reinforcement learning. Nature 518, 529–533. https://doi.org/10.1038/nature14236 In Agent Laboratory, Schmidgall and colleagues manage to replicate a Ph.D. student (in machine learning) using large language models and the Huggingface API. Given only a broad research question, the virtual student manages to perform and write up a relatively banal paper in under 2 hours for less than $20 (at publication a month ago – likely less today!). The article highlights, among other things, how AI can facilitate the academic research process and how AI systems can be implemented in practice. Schmidgall, S., Su, Y., Wang, Z., Sun, X., Wu, J., Yu, X., Liu, J., Liu, Z., Barsoum, E., 2025. Agent Laboratory: Using LLM Agents as Research Assistants. https://doi.org/10.48550/arXiv.2501.04227 Using similar methodology, Su et al. developed BioMaster, a system which promises to automate the bioinformaticians role. Using several role-playing agents (the Planning Agent, the Debugging Agent, etc.), the system is able to independently plan and run complex bioinformatic workflows. Su, H., Long, W., Zhang, Y., 2025. BioMaster: Multi-agent System for Automated Bioinformatics Analysis Workflow. https://doi.org/10.1101/2025.01.23.634608 Web articles Dario Amodei has a couple of interesting posts on his blog. One is a doomsday take on the soon-to-come unipolar/bipolar world layout shaped by advances in AI. Also includes a breakdown of the recent frenzy around DeepSeek. On a more positive note, in Machines of Loving Grace he discusses the areas where AI is likely to have the greatest impact (biology is at #1): https://darioamodei.com/on-deepseek-and-export-controls https://darioamodei.com/machines-of-loving-grace Simon Willison’s blog is a great source for AI news and commentary. For example, see Things we learned about Large Language Models in 2024: https://simonwillison.net/ https://simonwillison.net/2024/Dec/31/llms-in-2024/ The Matchbox Educable Noughts and Crosses Engine was one of the first machine learning algorithms. Implemented only using matchboxes (no computer needed!), it could learn how to play tic-tac-toe optimally. As a straightforward illustration of machine learning, it’s hard to beat. https://en.wikipedia.org/wiki/Matchbox_Educable_Noughts_and_Crosses_Engine Podcast episodes In a 5h-long interview with the brilliant creator of the AI assistant Claude, a philosopher/ethicist who trains Claude aligning it with humans (yes, it is an actual job), and a researcher who investigates mechanistic interpretability of AI (brings up parallels with neurobiology and what biology has learned from AI research and vice versa) – worth your time. Also available as an audio podcast: https://youtu.be/ugvHCXCOmm4?si=1PZLbKe6U1bF3ft7 Some relevant background information: https://darioamodei.com/ https://askell.io/ https://colah.github.io/about.html A deep dive into the reasons behind the most recent DeepSeek frenzy. Brace yourself for a lot of technical jargon and exciting discussions on what is relevant for AI: from breakthroughs in engineering multi-head latent attention of the algorithm to the unprecedented work ethics of Taiwanese producers of semiconductors: https://youtu.be/_1f-o0nqpEI?si=yIyJ7qKtkMyUqxeI The CEO of OpenAI, creator of ChatGPT talks about the greek-level drama of running a company that is at the edge of current dreams for humanity: https://youtu.be/jvqFAi7vkBc?si=AUuV7Nsi08jWBieV An interview with Roman Yampolsky an AI safety and security researcher who very casually proves his point that the AGI will be the certain end of humanity: https://youtu.be/NNr6gPelJ3E?si=89Y8CEKd4sR9KO0m An entertaining round table conversation with Nobel laureates of 2024, where they disagree with each other pretty much on every single point (including the questions about AI): https://youtu.be/1tELlYbO_U8?si=RnObU3HklfJ37TCu