有个比我年长挺多的mentor分享说，你如果想要更快的growth，就必须得选对赛道，选对你的团队；如果你所在的赛道或者团队不在高速增长，那么你的努力很有可能不会是等比例的回报。

这个道理其实感觉挺直观的。风口上猪都能飞，所以选对风口固然重要。只是之前可能没有这么直白的考虑过这个问题。

估计接下来还是要考虑赛道和团队的问题了。

My kids are adults now, but the big thing that I remember is that the days were long, but the years seemed short.

想想挺同意的，每天感觉都在照顾各种琐事很漫长，但是想必多年后回顾的时候会觉得时光飞逝吧。

其实前年2023我们组大org的头头离职的时候，我就很大的震惊。不理解为什么他要离开，不理解为什么他会放弃他的上升轨迹（事后看起来他去新公司从工资角度来说是个极其正确的决定）。当时正值暑假，忙着收尾一堆实习生的项目，而且家里当时也有一些个人琐事就没再多想。

去年2024其实分给我们组的资源就已经明显下滑。对接产品eng组也十分拉垮，我们使劲推进也没能在年底前完成。总体，去年太太在怀孕等待宝宝出生，也没心思折腾。又熬了一年。

今年终于迈出重要的一步，感觉宝宝没那么折腾了，想要认真对待自己的职业发展了。内部有一些机会，想要考虑的话，规定是得告知现有的经理，于是乎我也告知了。1:1的时候很紧张，因为没进行过这样的对话。

其实我并没有讨厌现在的组，只是我想寻找更加高速成长，影响力更大的机会。现在的组很舒服，人很好，但我总觉得缺了什么。

客观来讲，很多内部的机会估计都会不了了之，因为会招到比我合适的人。十有八九的几率，几个月后，我可能还是会在现在的组干着类似的活。尴尴尬尬的每一天接着干活。也不知道会不会因为暴露了我想换组而不被组里器重，如果发生也挺能理解的。

我其实不太确定我是不是做了正确的决定，但是直觉告诉我，是时候做一些改变了。哪怕短期会有很多阵痛和尴尬，我希望对我的长期规划能有到帮助。

温水煮青蛙，煮久了自己都失去了很多重要的能力。而2025年就希望我能慢慢的把这些能力找回来，看能不能有新的机会。

（不知道看我这帖子的几十人里有没有在招人的，有的话可以私信我，交个朋友聊聊。或者看到合适的机会的也可以告诉我😄）

https://microblog.lidingzeyu.com/i/personal-software-or-lee-robinson-6v33Zc03pQA/

Quote:

Creating software is starting to resemble cooking.

If you want a meal just for yourself—maybe you prefer your eggs a particular way—you can do that without needing a professional chef. The results might not compare to a top restaurant, but that’s not the point.

Your home-cooked software is exactly what you need, without extra fuss or cost. And each time you build something personal, you gain a deeper appreciation for the craft of creating software.

Does that mean frontend development or design is dead? No, not at all.

The more that build software, the more who will appreciate using great software. Cooking at home makes you appreciate going out to eat.

In his video he went over LLM foundation model training, supervised finetuning, and the latest exploratory RL and RLHF training. And there is a section where he explains why DeepSeek R1 paper is a big deal (at a high level, yet super helpful). I feel like I have a much better understanding of how RL is helping LLM post-training and how RLHF is (and is not) really RL and its shortcomings.

I watched the whole video (3.5hours @ 2x) and would recommend it anyone who’s interested in LLM or working with LLM since it will help you understand its quirks.

Note, I watched many of Andrej’s previous videos. For example, the one on tokenization (which explains why asking GPT to output YAML instead of JSON saves more tokens, among other things.) He is a great educator. 

https://mtlynch.io/notes/cline-is-mesmerizing/

In particular its opening paragraph (emphasis mine):

I tried out the Cline AI assistant yesterday, and then I went into a trance for five hours where I couldn’t do anything but stare transfixed at Cline fixing bugs for me.

And this part:

That’s when I was hooked

After Cline completed its initial implementation, I kept writing new testcases and then watching Cline update its code to satisfy my tests.

And that’s when I was hooked. I was so amazed that I could develop software this way. I just told the tool what I wanted, and it kept doing exactly what I asked.

I wonder how we can design video editing tool that can hook user to use it for hours? Going a bit broader, how do we design (traditionally) complex tools such that it's enjoyable and fun to use, like games.