Quality has dipped so much in last 2 weeks
retained text
For context I’m a university student outside the EU (re: I didn’t have the watermarking until recently) and I use Claude for pretty basic stuff — calendar management, asking basic content related questions, “grading” written assignments, editing lab reports, making flashcards, and helping with exam prep strategy. My setup is also pretty simple, each course has its own project where I include general preferences and instructions or notes for course-specific deliverables (e.g. this course teaches C++ so all code examples should be in C++ or lab reports for this course cannot exceed 5000 words + appendix). I haven’t spent too much time explaining specific instructions or specific preferences in detail, because I found in the past it wasn’t really required, though I might have to do so now because of the enshitification. I’ve noticed that Claude has become so bad at following instructions, and its context window and comprehension skills have significantly decreased as well. Here are some of the things I have noticed over the past 10-15 days, which weren’t really there for me prior: * Before, I didn’t need to think much about prompt engineering for simple tasks. “Explain what the deliverables are for this assignment” used to be enough. Now I need something like: “Explain what the deliverables are for this DNA sequencing lab. Split the space evenly between the lab notebook requirements and the lab questions. Explain how the code works and what needs changing to process the data correctly.” I don’t use Claude for complicated work, but even then, some days I spend more time writing and editing prompts than actually working. * One of the main instructions in all my projects is “Ask clarifying questions when relevant; ask questions before proceeding with a task unless absolutely certain.” It mostly ignores this. Worse, it answers first and tacks questions on at the end. Recently, I added “Ask clarifying questions before proceeding” to the end of a prompt, and it still made 700+ flashcards first and asked its questions after. I’d also flagged the instruction in the chat and in memory. It still asked at the end, so all the flashcards needed reworking multiple times, which used up a lot of my limit. * I don’t know if it’s the context window, comprehension, or both, but it forgets instructions almost immediately. I asked if my regrade email to my prof was easy to understand. He’s pretty anti-AI and wouldn’t appreciate an AI email. Claude critiqued my tone, called my request too detailed, and rewrote the whole thing, even though I have an instruction against rewriting my writing unless specifically directed in the prompt. I said I only needed to know what to change, that the detail is what a prof needs to consider a regrade, and that I didn’t want a rewrite. It apologized, then wrote a nearly identical slop email. These are just some of the examples that have pissed me off a lot lately, but I find myself having more and more of these experiences (up from maybe 1-2 times a month to nearly every time I use Claude semi-extensively). I’ve asked my dad who uses a lot of different AI tools for work if he feels the same way but he says that its much better than a few months ago and quality hasn’t really changed for him over the past 2 months. Maybe it’s because he uses API version and I’m just on pro but I really feel like I’m the only one because my friends don’t seem to notice a difference at all. Has anyone else been feeling like its really bad now or am I in a weird timeline rn?
document fetched 2026-11-11 00:20 · canonical https://reddit.com/r/ClaudeAI/comments/1xt1asu/quality_has_dipped_so_much_in_last_2_weeks