Aditya_kachhawa

Sorry for the late reply, I missed the notification earlier. What I'm really curious about is how modern AI systems work end-to-end in production. I can code and understand the basics, but I want to get into the real-world implementation details. Like, how do companies actually structure their training pipelines? What's the data collection process like at scale? How do they handle data quality issues? What are the actual trade-offs teams face when choosing between different approaches? I'm also interested in where these systems break down in practice and what that means for developers building with them.

I started a small blog about tech and marketing. Honestly, I got into this thinking about marketing, but I realized quickly that if you don't have an audience that trusts you, marketing doesn't work. The only way to build that is by creating genuinely valuable content. So now I'm focused on learning properly and sharing what I figure out.

Since I just wrote about ambient AI, I want to go deeper into how these systems actually get built in practice. I know "how AI works" is massive, so I'm thinking of starting with the data pipeline how training data gets collected, cleaned, and prepared at that scale. Feels like that's foundational but doesn't get nearly as much attention as model architectures.

Your question really helped me think about this more clearly. I honestly wasn't expecting anyone to respond, so I really appreciate you taking the time. Thank you sir!

Kind of wild to think the version control system I use every day was built before I could even walk. Really appreciate you sharing that firsthand context, you don’t usually get that kind of perspective.

I also found your Truth and Fascism piece on your site, and the correspondence vs. performative truth idea really clicked for me. The way it talks about institutions feels pretty universal — different forms, same power game, no matter the country.

Oh wow, you were following the mailing list when it happened? That must've been entertaining.

Did it feel significant at the time or just like another "Linus is pissed and building something" moment?

Google killing APIs has become so predictable there's literally a website tracking it (killedbygoogle.com).

What's wild is they give you 3 months to rip out and replace something that might be a core feature. That's not a migration timeline, that's an emergency. This is why vendor lock-in concerns aren't just paranoia anymore.

Was reading about version control history and found out Git went from first commit to self-hosting in like a week. Linus was just mad about the BitKeeper licensing thing and hammered it out. Not some grand architecture - just "screw this, I'll do it myself." And somehow that became... everything. Wild.