Week #111

It was a fun week. A rollercoaster to be honest, the weekend was depressing a good take off in the work, but a dip of sickness and back to work in the end. So, a lot happened this week, the gist is, I have a lot to learn, learning and exploring just in spite is still a thing that humans do and will continue doing, no matter how bad the AI situation gets. That’s a positive thing I am making myself believe from this week.

Quote of the Week

“Never measure the height of a mountain, until you have reached the top. Then you will see how low it was”

— Dag Hammarskjöld, Markings

Beautiful. You think the trek is 20 kilometers, that might be easy, but then you climb and after reaching you tell, man that felt like it was 25 kilometers. That last kilometer never actually completed. I have felt that, you cannot put a number on the beauty and the gigantic-ness of nature’s marvels. If you do, you’ll often undercount that and never realize the beauty it’s trying to show you. That’s the kind of adventure where I shut off my scientific-mathematical brain off. Otherwise it just doesn’t feel satisfying, the feeling of gratefulness is something I develop from that experience, and that is something you cannot put a number on it, or cannot appreciate if you put a number to it. Just live the life with an awe and don’t try to prove everything scientifically, sometimes there is beauty and things beyond science.

Read

  1. How to write with an LLM

    1. This is beautiful advice 1. You may not use a single word an LLM suggests to you. and 2. Avoid encouragement . This is so true.
    2. I use it all the time, even on this piece, it gives me a score each time on how good it is, with each time due to the memory thingy, it has gotten really encouraging and that I have started to refrain asking it on my account. I use a private chat conversation to rate and find errors in my newsletter or any piece of writing.
    3. Personalisation with LLMs is good at a certain level, after that it becomes garbage. But the point of this article is that you cannot outsource your thinking, you have to write the words out. Spill the beans, that’s what the world needs, that’s what you need.

  2. Learning programming in an age of LLMs

    1. I respect the people that have such experience, but it sets them a bit far from the reality and the expectations of the society in the modern world. With the amount of change happening in software or any field these days, taking advice or giving advice is a bit uncomfortable thing, nobody can predict what would happen.
    2. But I like these advice and piece. It shows, that you cannot really outsource creativity. You can show it for a while, but then you have to know what you are selling. If you don’t then you’re screwed.

  3. Martin Fowler: I don’t like LLMs

    1. Perfect. They find working and interacting with humans over a period of time satisfying and rewarding as opposed to working with an LLM. Who can argue with this point. I can’t and I won’t because there is no counter point for it.
    2. LLMs don’t have the care and emotion that a human can show while interacting, it can only glob about how good or bad the current situation or problem at hand is. Its just approximating how a human would write, nothing much.

  4. Bad code is Kudzu

    1. Ok, this is fair take, you have to make sure the plants don’t overgrow and intervene the useful ground. This is good analogy. You have to trim it, keep it short and grounded to the limits. Just like ai slop.
    2. You have to maneuver the codebase in such a way that it doesn’t collapse or make a mess for you or your future coworker.

  5. I am not addicted to my smartphone or social media, I am addicted to information, to thinking while consuming

    1. Eye opening take here, “However, my brain is still craving for those highs from the early internet. It craves for the discovery, the exploration” Brain carves information, the information that it has worked through, not spoon fed slop and macro level info dump.
    2. Stimulation of brain is what is making it hard to complete the satisfying loop. We are short-circuiting the brain’s reward with the short form content. We have to break it out, to allow it to get bored and then explore some ideas, connect a few dots and get you excited for the action.

Watched

  • Pedagogy with AI

    • A brilliant lecture or conversation. It highlights a question that really needs to be addressed “What should we or younger generation learn, if knowledge is no more the moat of humans”.
    • The points discussed here are a bit alarming, like students responded with we come to university or school for knowledge and understanding humans (socializing), if that is the case and LLMs can give the knowledge then is university’s only reason for existence is making students socialize? I don’t think so, there are more reasons like cultivating confidence, skills that just knowledge cannot, more importantly it gives humans a teacher, a teacher is not just who gives education but also one who understands the character of a human. LLMs cannot do that.
    • The other point is the economic thing. I think that would get worse for sure, because there is no control on how these models are deployed and used. So, it can cause massive social or civil wars for sure. That sounds dystopian but it might just be a couple of years away.
    • Indian societies should really create places to discuss such matters because Indian culture is different compared to western, we have a lot of knowledge in shastras, but deciphering the use of it in such times is what is need of the hour, I don’t think there is any problem that our culture cannot solve as a society if worked with harmony and selfless motive.

  • Is AI a good research assistant: Wading through AI - Episode 9

    • Yes, this is validating. LLMs cannot read between the lines. They cannot have human like humor, but humor aside, it cannot even fact-check itself. Yes, once you have the piece yourself, it can do specific things to check, but it cannot do something a human with curiosity and discipline can.
    • The Gemini banter is so good. Its in its own league of no league. I am not sure about GPT and Claude personality, they both have quirks. GPT’s reading style is a bit disappointing and Claude’s confidence is just jarring.

  • Planetscale: S3 Doesn’t work the way you think it does

    • This is fascinating. Creating the chunk and then having the other number based on the file data. Also a learning that S3 is not file system, its a key value store kind off. Like the key is what distinguishes each file or blob from other.

  • Please stop using stupid Models

    • This is very useful, I didn’t think it that way “raising the ceiling and raising the floor”. damm. that is so mathematical. Ceil is when we round up the number from 5.7 to 6, flooring is we round down the number , from 5.4 to 5 right? That just make sense now the wording I mean, raising the floor. wow
    • Other thing about models being a bit probabilistic over a larger tasks is correct, if you give small task, the reason to predict some things is small, the moment you take it larger, the probability will exponentially drop or increase based on the error or success rates right?

  • Did AI Kill React Native?

    • I didn’t think Agents can build apps reliably right? Maybe I haven’t experimented with right models just yet, but it doesn’t look good.
    • I didn’t knew React Native was a javascript embedded into the app that orchestrates the native components. Really fascinating. I like how it would update the app without updating from the app store, wow. Why don’t people do that? Bank apps are nightmare.

  • The Standup #72: TUI or GUI

    • That was a good one. Neovim GUI. I don’t want to use it, nor it was meant for. As Casey pointed, its not about usability.

Learnt

  • Actual difference between CPU and IO bounded task

Tech News

For more news, follow the Hackernewsletter (#810th edition), and for software development/coding articles, join daily.dev.

That’s it from the 111th Edition of techstructive weekly. I hope you found it helpful, and relaxing. If not please drop any suggestions, feedback or discussion about certain things you want to in the comments or drop me a message on my socials.

Thank you for reading,

Until next week.

Happy Coding :)