Dear all,<p>I am currently a Master student in Math interested in discrete math and theoretical computer science, and I have submitted PhD applications in these fields as well. However, recently as we have seen advances of reasoning capacity of foundational models, I'm also interested in pursuing *ML/LLM reasoning and mechanistic interpretability*, with goals such as <i>applying reasoning models to formalised math proofs (e.g., Lean)</i> and <i>understanding the theoretical foundations of neural networks and/or architectures</i>, such as the transformer.<p>If I really pursue a PhD in these directions, I may be torn between academic jobs and industry jobs, so I was wondering if you could help me with some questions:<p>1. I have learned here and elsewhere that AI research in academic institutions is really cutting-throat, or that PhD students would have to work hard (I'm not opposed to working hard, but to working <i>too</i> hard). <i>Or would you say that only engineering-focused research teams would be more like this, and the theory ones are more chill, relatively?</i><p>2. Other than academic research, if possible, I'm also interested in pursuing building business based on ML/DL/LLM. <i>From your experience and/or discussions with other people, do you think a PhD is more like something nice to have or a must-have in these scenarios? Or would you say that it depends on the nature of the business/product?</i> For instance, there's a weather forecast company that uses atmospheric foundational models, which I believe would require knowledge from both CS and atmospheric science.<p>Many thanks!