Indraneel

Software Architect, Organizer Azure Meetup Frankfurt, Teaching ML/AI at Frankfurt UAS, Writer / Blogger for various Marathi Newspapers and Magazines

[Indraneel @indraneelpole]

Won the Renault challenge @TechCrunch Hackathon @ Paris
@ddobric @daenet

[Indraneel @indraneelpole]

@sandesh_samant Repetitive, and they ride solely on nostalgia wave lately.
They need to take a break and come up with a better season 3 now

[Indraneel @indraneelpole]

@sandesh_samant There is in-built heating anyway. AC is useless after a few weeks every year. New constructions mostly have heat pump that also works as AC in summers, older ones don't however.

[Indraneel @indraneelpole]

Hits home close after buying my third kindle this exact way

[Gabbar @GabbbarSingh]

“There is some screen thing in your bag, with a battery” The guy at the bag-scan told me.

“No chance, I took out every piece of electronics. Maybe it’s the phone charger” I implore, but he isn’t satisfied. So I look for all possible pouches.

And then it comes, behind an obscure zip, My Kindle. Which I thought was lost, and I bought a new one, after sufficient months of mourning, and justifying reading books on your phone itself.

Thanks to the Lucknow Airport security guy.

[Satya Nadella @satyanadella]

https://t.co/vLmiBKTtX3

[a16z @a16z]

World Labs CEO Dr. Fei-Fei Li: "The world is not made of words."

"Language models have given machines an extraordinary command of concepts, vocabulary, and reasoning, but the physical world, virtual or real, runs on a different substrate."

"Where language models learn the statistical structure of text, world models learn the statistical structure of space and time: how light falls on a surface, how a garden looks from an angle no camera has captured, how objects respond to force and follow the laws of physics."

"Language gave machines a way to talk about that world. World models are how machines will finally come to understand, imagine, reason and interact with it."

Full piece: https://t.co/C9qOJg5wuc

[Indraneel @indraneelpole]

@DIBookkeeping This is good, I prefer BabySmash by @shanselman works offline

[Indraneel @indraneelpole]

In the age of AI, it is always good to get back to basics every now and then.
These two books, The Mythical Man-Month and Clean Code, shaped the foundation of my software engineering career.

Up until the age of 20, I didn’t know how to code.

That probably sounds dramatic today, but many from my generation had a similar journey. There was no ChatGPT, no endless YouTube tutorials, no AI pair programmer sitting beside you. Stack Overflow existed, but for beginners it often felt always more condenscending than welcoming.
During my undergraduate studies, I honestly didn’t pay much attention to programming classes. But at some point, reality hit me: if I truly wanted to learn software engineering, there was no shortcut.

So I did it the hard way.

I picked up a standard C#.NET book and, for nearly two months straight, I read, re-read, and typed every single line of code into my laptop.

Not copy-paste. Typing.
I broke things, I fixed them, I fought with compliation errors, build errors, syntax errors, and spent hours on finding missing semicolon.
That process taught me programming.

Later, I came across these two gems, and they completely changed how I thought about building software.

The Mythical Man-Month taught me that software engineering is fundamentally about communication, complexity, and managing human systems, not just about writing code. Brooks’ Law (“adding manpower to a late software project makes it later”) is still painfully relevant even in modern AI-driven development teams.

Clean Code taught me discipline:
Write code for humans first, machines second, because:
Naming matters
Small functions matter
Simplicity matters
Technical debt compounds faster than people think
Good software is crafted, not merely generated.

Even today, I revisit parts of these books from time to time to refresh the fundamentals.
Whether you are a “vibe coder” or not, whether you have a copilot or not, whether your agents write code for you or not, there is still no escaping the hard path if you truly want to become good at software engineering.
Tools might evolve, abstractions migh change, AI might take over.
But fundamentals remain stubbornly timeless.
#WeekendGyaan

[Indraneel @indraneelpole]

And so you need information about recent orders over API but the LLM would think it is better to tell you about invoices instead and thus you will recieve what the almighty LLM provides

[Yegor Bugayenko @yegor256]

RESTful APIs may be dead soon. Instead, web services may expose a single POST entry point for a prompt. Internally, an AI agent may decide how to interpret it and what to do with the data and the database.

[Indraneel @indraneelpole]

@benedictk__ This made to my timeline though

[Indraneel @indraneelpole]

@back2thevibe Deool

[TuNichtGut 🕊🍉 @M4GGIIIgesperrt]

Ich auch

[Indraneel @indraneelpole]

“Why don’t we just put mirrors in space and light up cities at night?” type of question

[Indraneel @indraneelpole]

64GB iPhone. lol

[Indraneel @indraneelpole]

Just don't.

[Indraneel @indraneelpole]

@girishkuber @croatia Plitvice?

[Indraneel @indraneelpole]

@GergelyOrosz I think self hosting and cooldowns are the two most practical approaches from this list. I wonder though, is this tooling specific, meaning npm and py packages are seeing more such attacks than nuget for example?

[Indraneel @indraneelpole]

So the real question is, Are there computational principles in cortical systems that could complement current AI architecture?

[Indraneel @indraneelpole]

In last year or two, I have been exploring ideas around contextual association and distributed learning architectures. Mostly that are inspired by cortical processing. One of the core aspects of cortical processing is that the brain doesn't appear to work as a single centralized system. Many regions process information in parallel while continuously exchanging signlas related to context, memory, prediction, sensory inputs etc.

Now look at neural networks, including transformers, they still largely trained as end to end monolithic systems with globally optimized representations. They are very powerful at pattern extraction and sequence modeling, but underlying computational structure is still quite different from how distributed cortical systems are believed to operate.

It is interesting to see newer research directions gradually revisiting ideas around modularity, contextual coordination, heterarchical processing, and distributed world models again.

[Indraneel @indraneelpole]

For example, humans are very good at continual learning without catastrophic forgetting. Or, contextual reasoning. Or, transferring knowledge across domain. Or, integrating multiple sensory modalities.

Current deep learning systems still struggle in some of these areas. Or, they require enormous compute/data to approximate them.