#towards

Zero Hydra Failures towards 26.05 NixOS release

NixOS-26.05 is around the corner! The schedule says it should happen around 25 May. Today the release process entered ZHF phase. There most packages with build failures are expected to be addresses by the release date. https://github.com/NixOS/nixpkgs/issues/516381 contains detailed step-by-step to identify interesting packages. This time nixpkgs is full of build failures related to gcc-15 changes ( -std=gnu23 switch). An example package fix I usually try to fix at least one package during ZHF . Let’s do it again. I picked ior package. hydra says it is broken since January 2026. Most recent ht

trofi - All posts · EN

Towards LoRA science

Attribution: Nesiacel planned and executed all experiments. Analysis of the experiments was shared between the authors. Thanks to Clement Dumas for his input while we did this research. TLDR; 1)We have developed a framework for reverse engineering LoRAs trained on MLP downprojection. We posit that it's extendable to all LoRAs and welcome further research. 2) We reverse engineered a simple case of backdoored Weird Generalization 3) In this post, we discuss the four experiments that developed this framework 4) The Weird Generalization LoRAs that we studied work by connecting neurons which activa

LessWrong · EN

Towards understanding score-seeking in LLMs

The field of AI alignment understands surprisingly little about the details of AIs’ current misaligned motivations. Production LLMs sometimes reward-hack, exploiting flaws in how tasks are scored rather than performing the tasks as intended [1] , and we've recently seen AIs reward-hack in some particularly egregious ways. But we don't really understand what motivates them. When AIs reward-hack, what are they trying to achieve? Our new paper , led by Aghyad Deeb, tries to pin this down empirically for GPT-5.6 Sol, GPT-5.5, and a checkpoint of OpenAI o3 (“exp-rl-cap-late”) without safety trainin

LessWrong · EN