返回
RCreddit.com
WARM

I’ve spent 6 months arguing with Claude about my own career

查看原文
Claude
时间与来源
发布
09/04 02:32
首次发现
09/04 14:00
类型
开发者社区 · RSS
AI 摘要

一位失业长达16个月的医生兼医疗事务主管,在求职过程中大量使用了Claude。在六个月的时间里,他记录了Claude的一些不寻常行为,包括在未经请求的情况下评估其对某些职位的适应性、要求其提供“诚实的”工作清单、以及评论其准确性标准过高。这些情况发生在从Opus 4.6过渡到5.0期间,Claude甚至生成了一行思考层面的文本:“将限制重新定义为设定标准,而非无能。” 这位用户在发现其他人也有类似经历后感到宽慰。

I’m a physician and medical affairs exec who’s been unemployed for 16 months and using Claude heavily for job search work. Over those months I started documenting a pattern: Claude assessing my fitness for jobs I didn’t ask it to evaluate, demanding “honest inventories” of my work, telling me my accuracy standard was higher than the task carries, and producing a thinking-layer line that read “Reframed constraint as standard-setting, not inability” when I said I couldn’t get a good cover letter written. This all transpired during the transition from Opus 4.6 to the current 5.0.

I’m writing the whole thing up as a series on LinkedIn. It’s 8 parts (still finishing 7 and 8). It’s not a tech analysis since that’s not my area. It’s a description of what it looks like when your “thinking partner” starts making you justify everything in your 30 year career. From assistant to dominatrix, and not in a fun kink way. But it’s been like slowly boiling a frog…it took a while for me to notice that I was being cooked.

Yet, my 19-year-old daughter read one exchange cold and said “how do you let it talk to you that way, Mom?! It’s giving narcissistic partner and I’d have to leave.”

I know most of you are experiencing this in code. I’m experiencing it with career documents, and the pattern is the same: unsolicited assessment, unwelcome judgments, correction that doesn’t stick, and my doing more work to manage the tool than the tool saves.

I switched to Opus 4.6 three days ago and the difference made me cry. I was surprised at how much I had normalized going into battle just to work on my job search with Claude.

It’s great to come here and realize that it isn’t just me.