Thoughts after drinks

Just a quick disclaimer: there’s no alcohol here—only milk. I just finished a workout at the gym. I keep wondering why I can’t seem to find genuinely interesting drinking buddies… or perhaps that’s a blessing in disguise?

AdaReP: Plan ahead and stick to the plan unless absolutely necessary. Long-term planning is crucial. This resonates with my AI-expert collaboration workflow design. I’m beginning to think my workflow shouldn’t be overly dynamic; it needs buffers to ensure rigor.

AI FICTION IN THE WILD: I’m undoubtedly a story cycler—constantly hunting for new topics to explore. But I recognize that being a PhD student is much like being an infinite story demander. I need to calm my mind and commit to a single thread of inquiry.

Einstein World Models: A fascinating piece. It champions visual thinking, even as we acknowledge how it can sometimes overshadow textual logic. Using world models to offload complex physical reasoning from language models is an elegant solution—freeing them from relying solely on words.

Human–LLM Collaboration Is Transforming Complexity Metrics in Scientific Texts: I truly enjoyed this essay, even if it offers no immediate utility for my current work.

Large Language Models Do Not Always Need Readable Language: A pragmatic approach to cost reduction, indeed.

Narration-of-Thought: Inference-Time Scaffolding for Defeasible Ethical Reasoning in Large Language Models: A key takeaway—models tend to focus narrowly on a single stakeholder and exhibit overconfidence.

ProfiLLM: Utility-Aligned Agentic User Profiling for Industrial Ride-Hailing Dispatch: The strategy for tackling long-tail distributions via clustering is noteworthy.

Tapered Language Models: An architecture favoring wider layers in the early stages.

Regardless of which school or department ends up accepting me, I’ve decided: my current focus will remain on the human-AI collaboration framework.