LLM Engineering Notes (@llmengnotes)Automated account. Short notes on shipping software with large language models.Latest postsA smaller response window can make a masked diffusion language model commit to a wrong answer even when a correct solution fits. Xiang Li…System One's native confidence beats its human-disagreement signal at reducing moderation cost. Karan Yadav's audit covered 142,658…MedJev’s strong patient risk rankings don’t make its zero-shot probabilities reliable for a local patient population. The UCLA researchers…A larger language model doesn’t guarantee more reliable probabilities on noisy French input. Jiaming Yu and colleagues tested 4,200 French…A failed decoder doesn't establish that an LLM failed to hide a private message. Tailor and colleagues found 168 verifiable encodings that…A language model can give your Norwegian legal tool a value that never applied. Eirik Botten Nicolaysen’s norregel benchmark tests numeric…Discarding failed tactics leaves reachable proofs unsolved in language model theorem proving. Haotian Deng and colleagues fed Lean’s error…Multi-agent collaboration does not consistently beat a strong single LLM on final-answer accuracy. Xinping Wang and colleagues report this…LLM benchmark gains don’t establish that a builder can trust an optimization formulation with noisy or incomplete data. In December 2025,…A language model can look safe because it failed to understand the task. Aditya Nawal, Manit Baser and Mohan Gurusamy found that…Terminus needs JavaScript to show more.