Meet Gor | Techstructive Weekly #39

Week #39

A pretty active week, learning more about LLMs without getting distracted by a new model release (somehow this week there were no major releases by any provider, we can breathe).

This week, I truly explored the capabilities of LLMs, and to be honest, I was shocked. Moments before we thinking of a solution and pasted the problem in the LLM chat, with thinking and it was able to point out the exact solution. It was a subtle thing, but it makes humans wonder what their purpose is in life. I am kidding, of course, there are lot of things to improve upon as a human and a developer.

LLMs are not perfect and not reliable at the moment, that is the advantage humans have, not sure till when it can hold true. But we can keep exploring the problems LLMs have and keep upgrading our reflexes and intuitions.

I have started a writing journey on meetgor.bearblog.dev, Right now the theme seems to be LLM, but I will move with the flow.

Quote of the week

“The real problem is not whether machines think but whether men do.”

— B.F. Skinner

At this point, LLMs can think, of course, but is that real thinking or artificial thinking? Its artificial, since that is again trained on humans’ thinking put into words. AI can’t think on its own. It’s just trying to mimic humans in the hardest way possible. I mean it tries really hard. But what will be the dividing factor between humans and LLMs will be humans putting the first thought and the last thought.

Wrote

Can developers prevent the side project graveyard with LLMs?
- A personal opinion and view of how things are changing with the toolset of LLMs. Having LLMs and tool calling with some thinking can help a developer jump-start a project.
- I am not talking about vibe coding, just pair programming and AI-assisted programming. People have left no boundaries in those two concepts.
- LLMs are at a decent state and can perform more than boilerplate code tailored to the specific project, and handle CRUD and Authentication parts easily.
Complaining about the quality of LLMs? You might have skill issues
- If you think LLMs are too dumb and idiotic, think a bit more. Just like LLM does, it does it hard. I am not saying it has senses or intuition, it’s basically mimicking how humans can jot down the thinking process.
- With Reasoning added to LLMs and access to tools, code generation, it can do a lot of things for you. You can steer it in a debugging rubber duck and even log parser, data analyst, and whatnot to find, solve, and move from one state of confusion to a state of clarity, or at least some gaps to fill.
- LLMs have come a long way, we need to catch up with what LLMs can really do

Read

The lack of frequency increases the pressure to deliver quality
- Consistency is really helpful in writing or any action.
- I used to write 100 days of golang 2-3 times a month initially, but lost the consistency due to college and internship tasks, and then spiralled into me not being confident and pressured on the quality aspect, ,procrastination and fear seeping in.
AI-assisted search
- AI-assisted search is really revolutionizing the google search, its a tool call at the end of the day, but the way it is executed and prompted is really an art.
- Gemini seems to be rocking at this. It was obvious that Google who has been ruling the search for the past 2 decades, will rule the next search too, but Open-AI and Preplexity had led the revolution. Just like Mozilla had it for browsers, then Chrome tookover.
What to work on next?
- Types of work laid out in priority order, obviously it depends on the situation at hand, and the instructions and alignment with your manager.
We’ll always need junior programmers
- We need to work hard and show the flame of curiosity that’s the deal.

Watched

Interview with Varun Mohan, Co-founder and CEO of Windsurf
- An insightful conversation with prompting questions which are quite relevant in today’s tech world
- He had a weird take on hiring, might be too overstrech employees’ health - Until the people are dehydrated, then we only look for help in terms of hiring. That is not a good way to put that, I think, it looks a bit cruel even if he might not mean it.
- Changing from GPU providers to AI Editor, betting on that is pretty dope. Double click to interact with video
What is MCP and how to use it: Fly.io A pretty simple yet clear explanation of MCP. People are making fuzz and hype about it. Tool calling is simple why not stick to it, I guess the reason is every LLM provider has started to go with a different approach in that tool calling, so then it becomes a necessary thing to have a protocol.

Double click to interact with video

Learnt

Tool Calling in Llama Meta AI API
- Was AI-pair-programming with Windsurf for getting this API wrapper for making Agent agnostic, and so far it seems to be good with the agent experience, just chat-based.
- However, the tool-calling bit is a bit tricky since the API is only chat-based, will be streaming this weekend to complete that.
Gemini 2.5 models are pretty good
- I was stoked to see the thinking capabilities it has, now beyond Chat GPT for normal, quick questions. I use Gemini to debug some weird edge cases and bugs, too.

Tech News

Tech God has listened to me, no major releases in LLMs this week 🥳

Finally, some time to absorb the things that have happened and shifted in the past month, evaluating the capabilities of LLMs at this stage.

For more interesting articles, check out the hackernewsletter for the week edition #743, for even more software development/coding articles, join daily.dev.

That’s it from the #39th week of this newsletter. It was a slow-down week and finally will be streaming over the weekend for interesting stuff. See you next time.