Week #35
It has been quite a busy week, with some ups and downs at work. Another Friday release, a lot of patching and end-moment clutches to get it on time. I am not bragging, it is not good, but it is unavoidable at times, I am so proud that I was able to deliver it, despite having a weird situation in life.
Life is tough, there was a rough day this week in my life, things just get worse and worse, and then get a little hope. That hope might be too tiny, not enough to heal the mind, but enough to carry life. It is tough, I need to stand and stay strong and find a way out because I am destined to be great.
I am receiving signs from the universe about the direction to go, but no matter how hard I try, even from the worse than me, they get results, but for me it just sucks, no response at all. Look at the post I created this week, 0 reactions, man! I put effort and it is not AI-generated, raw human emotion. Tough luck, but someone has said to do the job and not to focus on the result. That is what I am doing and leaving it as is.
Soon, it will change, soon I will be in the bliss of life.
Quote of the week
Robots generate, Humans manifest.
This is a quote I am making myself realize every time I have a problem to solve. I am a software developer, I get bugs, surely LLMs will generate some code, but I want to manifest the code, there is a difference. Humans can show things with emotion and work, AIs can’t do that, humans have feelings, but LLMs won’t have them.
Manifestation is far from generation, it needs a thought, it needs a raw emotion. Writing is not art, it is menial work, crafting stories is art; Generating images is not art, capturing memories, emotion, and expressions is an art; Speaking is not art, connecting, and resonating with emotion, is art; Writing code is not art, solving problems, expressing ideas is. You are a human, whatever you will create will be art. Keep crafting, keep exploring.
Created
I wrote 1 article, published 1 video (finally after almost 2 months), and 2 live streams.
Wrote
-
A letter for someone who is afraid of AI (including myself)
-
I focused on Manifestation vs Generation
-
Robots will generate, and Humans will manifest
-
Humans need to keep getting good at the craft, leverage AI to make it faster and more efficient
-
Recorded
-
SmolAgents: Simple and Lightweight Library for creating AI Agents in Python
I learned a lot, created a few examples-
Basic tools: Multiply and add (call me silly, but LLMs can’t add)
-
Specific tools: Scrape a website and list all the links and content
-
Answering a user query from a DB
-
Generating the SQL query from Natural language
-
Executing the SQL query as a tool
-
-
Using Local LLMs (llama3.2:1b and Gemma3:1b)
-
Using Cloud LLMs (Mistral Small)
-
Streamed
-
Porting my Blog to burrow SSG + CMS
Made some tweaks and refinements in the UI and finally, it is looking good. -
Improving the burrow CMS
Was brainstorming on how to make the edit and delete functionality in the CMS i.e. the sync between the Git and Database
Read
-
Your manager should never have to ping you: This post discusses that a junior should be asking questions, updating too frequently, or overdoing it, I get that point. The author is trying to make sure the manager is never left with suspicion or doubt about the junior “is he working on something?”, “is he even working?”, or “Did he understand the problem right?”, because time is precious, and no one wants to waste time, and resources in rabbit holes. Rabbit holes are great but not always and too much of them can deviate you from the problems in a broken way. So I agree with this advice, however, think before dropping the message, don’t try to spam, avoid jargon and I think the communication needs to be to the point. I sometimes get too nerdy or too off-topic in some discussions, that we have to get on a call, yes that happens. No one is perfect, everyone is figuring out life.
-
Learn to Code: Kache’s Twitter (X) article: This just blew my mind. If someone is optimistic and pessimistic about AI, this guy is just on the extreme end of optimism. Nothing wrong with it. Every coin has two sides, no one can perfectly balance both. But I was kind of doing that, and that was leading me nowhere, after reading this, I am nudged onto the optimistic side. Like I feel there is too much to be done for me, so little time. It just pumped me up.
-
The worst programmer I know: Yes, this is true. Some engineers or programmers do not actually work on something, but they make something work together. They are the hand-holders, the captains, the leads (not the tech lead, the leaders). That is one type of personality, and that is no right or wrong personality. Humans just are what they are, that’s what makes them humans. This brings me to another point, pair programming, oh! That is a charm, a secret little thing that somehow works wonders. Pair programming is a must, no matter what the scale of the organization is, it opens up the programmer’s mind in different ways, changes the perspectives of individuals, gives confidence, helps learn communication, eases the stress a bit, and makes programming more fun. I can go on, but that is an understatement for pair programming.
-
My 6 months with the GoTH stack: building frontends with Go, HTML, and a little duct tape: I have heavily done some Golang with HTML for my burrow SSG and CMS. I think at some point, I will have to leverage Templ and Tailwind, I am not a frontend guy, so you know. Templating is really a thing, that when you code in python doesn’t feel but in Go, it actually helps you understand what the heck is going on.
![](https://substackcdn.com/image/fetch/w_56,c_limit,f_auto,q_auto:good,fl_progressive:steep/https%3A%2F%2Fsubstack-post-media.s3.amazonaws.com%2Fpublic%2Fimages%2F7c4d69f6-05b5-48c1-9925-4938793dad50_512x512.png align="left")
My 6 months with the GoTH stack: building front-ends with Go, HTML and a little duct tape
-
Deep vs Shallow Go Interfaces: Deep interfaces are really easy and flexible to understand which is really a good observation. The shallow interfaces are the ones that are quirky but unavoidable, we need to have some level of basement laid in order to make it work, otherwise,it will be all deep interfaces with too many implementation trees. That I don’t like.
-
We should have more friction in our lives: Cassidy Williams Blog: I shared this with my own thoughts, and yes, we need to bring back some friction as well fiction (pun intended).
Watched
-
The Primeagen on the Lex Fridman Podcast
Man, that person is a legend.
I have only watched 1:30 hours of this, that is a 5-hour video, man!-
Addiction is something that feels like giving you satisfaction, but it never will, that is a great way to put it.
-
Work Harder, not smarter! This is the new advice (instead of working smarter, not harder) Why? Because, how would you know something is the optimal way to do it, before doing it? Skewed
-
-
Why is Google winning the AI race?
-
Google has controlled the internet for the past 2 decades, and it has Data, some model science, and also now has some infrastructure (hardware). It will control the AI race, Apple is actually surprising in all of them but dominates none of them.
-
Can Apple make a comeback? Looks hard, but we have seen comebacks (Google was struggling with Bard, and now it is toppling Open AI)
-
-
- That’s pretty interesting, git scrapper-based newspaper, wow, I want to make this, I want to aggregate my favorite blogs and platforms to get the most up-to-date content. Content these days is overloaded.
-
Will YouTube ever run out of YouTube Ids?
- Wow, YouTube is well thought out, 64 letters, that is a gigantic number, practically it will never run out of IDs. However what if someone spams a few billion video uploads?? That can pretty quickly bring the limit close, I mean YouTube will surely block it, but maybe who knows
Learnt
-
Python’s wired Pythonic expression: This I saw from Tsoding’s tweet
-
The expression was
70 * 2, 3 -
Expected is `140, 3`
-
But why sometimes, my brain thinks it should be 140, 210
-
Maybe that is too Pythonic, but that is not a tuple expression, It is just a singular expression, where there are two values evaluated
-
However, if you do something like this, this makes it more sensible
> 70 * (2,3) (2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3, 2, 3) -
Creating the tuple `(2,3)` 70 times
-
Wired! Sometimes, you don’t have to overthink, just make the thinking a little grounded, there’s too many nuances to keep in the cognitive stack I think.
-
Also, this makes it more clear, I think
> 70, 2 * 3, 5 70, 6, 5
-
-
Using smolagents with CodeAgent and ToolCalling agent
I was able to figure out what is the CodeAgent and ToolCalling Agent-
I was trying with the local LLM models those were pretty slow and kind of small (1 billion parameters) compared to the actual 100 billion and god knows how many billion is GPT
-
Then I tried mistral on the cloud, generated an API key and everything started to work.
-
The code agent is for generating code, and not explicitly for calling tools
-
There is a separate tool calling agent for that named ToolCalling Agent
-
I tried both of these for their particular use cases
-
Tool calling agent for simple predefined things like scraping the website and getting the links, getting the content, multiplying, and add (yes LLMs can’t do that for large numbers)
-
Code Agent for generating SQL queries based on the natural language for the query.
-
That was a great experience, this is what I love weekends for, trying, failing, failing hard, trying something new, trying hard, and success. Finally a new error message.
-
-
-
Using default dict with some wired default expression
profile = defaultdict(lambda: {"name": "", "links": []})I didn’t know we could use a lambda expression to create this default value for any key in a defaultdict, I thought that was kind of limited with the built-in datatypes like dict, list, int, boolean, etc. But this opens a wider door for nested data structures. Safe access!
Tech News
A lot of AI News, AI is no more hype, it is a revolution, a moment of life for most of us.
AI News:
-
OpenAI might have just fixed image generation with the new 4o Model
-
Claude shares the research behind the thinking capability of the chat model
Can we breathe for a moment, like there is hardly any time left to check out the releases? At this point, the only way is to adapt and keep struggling and learning new things. Keep the curiosity firing.
For more interesting articles, check out the hackernewsletter for the week edition #339, for even more software development/coding articles, join daily.dev.
That’s it from the last week of March, the end of quarter 1 of 2025. It has been a quick-flowing time, AI is taking over quicker than anticipated, and the pace is quite rapid. In the blink of an eye, a new model, a new improvement, a new tool, is getting launched. Times are wild, and so will the effort.
Keep the curiosity flickering!
Happy Coding :)
Thanks for reading Techstructive Weekly! This post is public so feel free to share it.
Thanks for reading Techstructive Weekly! Subscribe for free to receive new posts and support my work.