Joseph Bejjani · BlogWriting by Joseph Bejjani on AI alignment, reinforcement learning, and language models.https://josephbejjani.com/Can AI Write New War & Peace? A ‘Novel’ Without End in a World of LLMshttps://drive.google.com/file/d/1YLREG4HTe2M99jvLGHPrFv9z-y90MWSG/view?usp=sharinghttps://drive.google.com/file/d/1YLREG4HTe2M99jvLGHPrFv9z-y90MWSG/view?usp=sharingA final paper for Slavic 118: prompting an LLM to write new chapters of War and Peace at two junctures in the novel, and examining how its chapters and self-reflections differ from Tolstoy.Sun, 10 May 2026 00:00:00 GMTInoculating Language Models Against Misalignmenthttps://josephbejjani.com/blog/misalignment-inoculation/https://josephbejjani.com/blog/misalignment-inoculation/Inoculation prompting can mitigate emergent misalignment but may also create backdoor triggers.Fri, 30 Jan 2026 00:00:00 GMTDo LLMs understand their adversarial prompts?https://josephbejjani.com/blog/perplexing-poem-prompt/https://josephbejjani.com/blog/perplexing-poem-prompt/Discovering perplexing prompts that generate poems—then asking the LLM to explain.Mon, 22 Dec 2025 00:00:00 GMTWhen Agents Prefer Hacking To Failure: Evaluating Misalignment Under Pressurehttps://www.lesswrong.com/posts/AJANBeJb2p39su6F9/cs2881r-week-8-when-agents-prefer-hacking-to-failurehttps://www.lesswrong.com/posts/AJANBeJb2p39su6F9/cs2881r-week-8-when-agents-prefer-hacking-to-failureWhat do agents do when they face obstacles to a goal? If the only path to a goal requires misaligned action, will they choose it or accept failure? We build off Anthropic's work on Agentic Misalignment to investigate these questions in an agentic coding environment.Sun, 09 Nov 2025 00:00:00 GMTWhat I've learned doing RL with JAXhttps://josephbejjani.com/blog/mechagogue-jax/https://josephbejjani.com/blog/mechagogue-jax/Some of my experiences while working on mechagogue, a reinforcement learning repository with from-scratch JAX implementations of classic RL algorithms.Wed, 25 Jun 2025 00:00:00 GMT