Meet Gor | Techstructive Weekly #32

Week #32

Weeks of chaos have left me on the brink of burnout, with no time to breathe or reflect. Didn’t get much time to just sit and think, just had to complete a few hacks here and there to get the demo running, poor management and blind eyeing.

I am a bit stressed at the moment, not able to do anything, occupied in work from 9 to 12. Yes, that is a bit stressful and hectic to carry this daily.

Just raw thoughts in the start, feels a bit embarrassing but it is the reality. Have not uploaded a video since 4 weeks, will be 5th week this weekend if not uploaded. I am not worried about the number but I am trying to say how life is playing with me currently.

Too bad to be true.

I learnt a lot, not saying no to that, but at what cost? My entire week? This is what I do 24x7 apart from 8 hours of sleep? Maybe not.Its hard to answer and take a decision at the moment, since my only responsibility is to lift myself out of some situation.

Enough ramble, I did a lot of llm prompts, python all over.

Quote of the week

"The real problem is not whether machines think but whether men do."
– B.F. Skinner

At this juncture, I might not think, but I should have context on what to do. Figuring out the actual problem from the fuss, and then boiling that down to simple technical problems is something I need to work on. I realized this a lot over the past couple of weeks.

LLMs can think provided the human had did a bit of thinking at first. It is just going to predict the next word or code, based on what is already there in the internet, however some things are created of pure intuition and wisdom that is the one that will set someone apart from others.

Read

Why is everyone trying to replace software engineers: I had my first meeting where there were clients, actual users of the product. And I was a technical guy so not much needed for me to talk just demo the things, went a bit technical and in a jiffy but thats it, I think we will be replaced with the people who can understand the business better and not by any LLM. True.
Beware of the Makefile Effect: Adding already solved problem to your code is not bad, but randomly yanking and making a saphagetti of the codes surely is a problem. But the counter point to that would be, how to know if the thing we are copying is actually good, that sense need to be developed and I think that comes with time and experience and mistakes.

Watched

Really didn’t had much time to watch anything. On saturday and sunday though I watched the How to use LLMs by Andrej Karapathy. I mentioned this last week as well but this week I watched it completely.

ChatGPT Interpreter
Grok AI unhinged mode
Custom GPTs
Internet Search Tool

This was really insightful, I tried a lot of things, and one thing or the other really helped me solve so many intricate bugs and discrepancies.

Double click to interact with video

Learnt

We can use ChatGPT to debug the code, just provide the source file and could run and understand what to do and figure out issues
Also the interpreter mode can be used to provide a data source file and allow it to compute a few things to cross-validate with our code, this really saved me hours of debugging and quickly point out the issues, here is one conversation where the numbers were slightly off from my code and it helped me figure out that that was actually an issue in the data source and not in my code.
https://chatgpt.com/share/67c93174-2eb8-8011-b23c-e9651dffa91e
Providing a larger system prompt is powerful and we could cramp a lot of boilerplate data or any data source that instead goes into user prompt to save tokens. This is really powerful if we are making one shot calls, and not a chat like conversation systems.

Thoughts

Just a thought that might give you some insights and can make you think about somethings in a different way.

Writing LLM Prompts is the new writing code, we need to spend time thinking about different approaches for prompt and less on the code, as LLM would generate the code.

This is just my thought, I think I could write a article on it.
Next week I have 3 articles in mind

Use positive conditions
Python is tactile and golang is mechanical
Prompt is the new code, worry about the art of prompt rather than code

Tech News

Mistral OCR: This LLM things are taking over everythign slowly, every domain of things will now have LLM-equivalent, what do we call it? LLMfying, LLMing, GPTing, AIng, LLMify, ok enough jokes, but this is seriously good.
X let’s you query grok by mentioning it in reply: That is wild, I just got access to GrokAI App on my phone and its blazingly fast. Really that LLM is different from others, not much but a little open.

For more news, follow the Hackernewsletter and for software development/coding articles, join daily.dev.

That’s it from this 32nd edition of my weekly learning, I hope you enjoyed it, and leave comments on what you think about some of my takes or any feedback.