Undergraduate? No. We've had calculators able to solve undergraduate problems for decades. AI doesn't change the need to understand how calculus works any more than calculators did. The foundations remain valuable.
90% of the final grade are in room examinations with proctors, maybe two sets of exams of midterms and finals that the vast majority of the final grade comes from. This is already how most of East and South Asia does it anyways and it’s probably the best.
For publications and theses, as long as the final results hold and can be replicated and validated, I don’t see why we shouldn’t allow the wholesale use of LLMs
> 90% of the final grade are in room examinations with proctors, maybe two sets of exams of midterms and finals that the vast majority of the final grade comes from.
This is really just a glorified undergraduate education, the real point of graduate school is to learn to do real-world relevant research. For the latter, I think LLM use will be accepted but there will be a heavy expectation on the author of making the result very easily digestable for human mathematicians and linking it thoroughly with the existing literature - something that LLMs are very much not successful at, but a student might be able to do quite well with a mixture of expert guidance and personal effort.
I don’t think it’s just mathematics. We don’t hear enough about this, but if I think back to my undergraduate years, which were less than 10 years ago, every homework assignment and every take-home exam I had would be trivial for LLMs to solve at this point I wonder what is actually happening on the ground.
Well... here's something from "boots on the ground": I teach a bachelor's degree where programming is a smallish facet of a curriculum. My course is the last of a series of 3 courses which progressively introduce more concepts and try make practical implementations more feasible. I've been able to grade the course purely based on returns to take-home exercises, some of which are complex, some trivial. When ChatGPT (& Co.) came along I was still able to do that but with a major added workload to me (suddenly everyone started producing mountains of code, often nonsensical, but I still had to read it all). I always requested targeted, atomic changes to code (vs. rewrites) which served me well up to a point (I was still able to grade fairly). I requested them originally to avoid "github copies", but that worked kind of OK with ChatGPT too. However, when ClaudeCode came along it was obvious to me I'm loosing the battle. It does not particularly matter to me whether students use AI or not as long as the rows they add and alter in the assignments make sense, but the "last nail to the coffin" problem now with ClaudeCode is that in the latest batch (this spring) it is clear some students "pay themselves" a good grade (i.e. they pay for ClaudeCode, thus bypassing the need to actually learn). I cannot make assignments that are both complex enough to cause ClaudeCode tripping on something and still humane for those who do not use AI or only use free chatbot options. Essentially ClaudeCode plays havoc with the whole grading process: students not using it (whether they try to write code fully manually or ChatGPT assisted) are left with far less points that students who just push all the code I give to ClaudeCode and "let it rip" for some 15 minutes. This really irks me. So, my solution? Still working on it and hoping to find one! For sure no more points from most take-home assignments: lowest grades still achievable through them (the trivial ones), but that's it, the rest it preparation for an exam. Practically this already means anyone with ChatGPT is going to pass, no doubt about it... As for the higher grades, for autumn I'm desperately now figuring out how to even make a meaningful paper based exam for my course. I've myself completed a master's degree writing C language on paper with a pencil. I sure did not want to start doing that to others, but here we are. Besides, back in my youth the only "library" was pretty much ANSI-parts-of-C! I'm not sure what kind of a 2 inch thick stack of papers I'd have to give my students into the exam these days as reference material. One horrible aspect is that students are now far more dependent on compiler errors to spot pretty much anything and everything... I worry the first paper exam from me will be a total horror story to us all. In any case, interesting times.
I had this exact problem and came to the same conclusion. But for the exam, give them code and ask what it does, or give broken code and ask where the error is. Waaay less marking. I only asked for 3 small functions written by hand and that was still 90% of the effort to mark. But the marks felt valid in the end, so the process seemed to work.
I think the perfect combo is a powerful Mac Mini plus a portable iPad — let macOS handle the productivity work, and just use the iPad as a big-screen AI terminal.
I believe this is mainly due to China's strong sense of crisis regarding energy supply. If it had as abundant fossil fuel resources as the United States and Russia, it likely wouldn't have invested so heavily in diversifying its energy sources. The development and support of new energy vehicles can also be largely attributed to this.
Heavy dependence on coal has led to severe environmental pollution. As winter approaches and cold fronts sweep down from the north, increased heating needs drive up thermal power generation, releasing more pollutants into the air. During periods of stagnant weather with no wind or rain, smog settles over the sprawling plains, triggering both public health concerns and mounting pressure on the government to address the issue and explore alternative energy sources.
