What Do AI, Retiring Chips & Linux Packets Have in Common?

Hello World, and welcome to the new edition of The Scholar Diaries. Like always first lemme give you a gist of what is this all about - My name is Prince Bhardwaj, a researcher focusing on anything and everything about computer science. Most of my days are filled with chasing answers, running experiments, debugging code, and occasionally banging my head against the wall(the joys of research!). This newsletter is my "dear diary" way of sharing the interesting bits I come across during my work. Each week, I’ll post highlights from my notes like the ideas, discoveries, research papers, breakthroughs, and even the occasional head-scratcher that kept me up at night. You’ll find my take on where things are headed, insights from the papers I’m reading, and my thoughts on the open questions in computer science. Whether you’re a fellow researcher, a tech enthusiast, or just curious about what goes on in the CS (and no, it’s not just about AI), I hope you'll find value and maybe even a little inspiration here.

(Oh, and if you’re curious about more details about me, my work, or what makes me tick, you can check out HERE)

The Great AI Tug-of-War

Right now, the tech world feels like a high-stakes poker game. On one side: pessimists who think AI peaked with GPT-3.5 and now we’re just polishing chatbots tuning them into AI agents. On the other: optimists betting their life savings that AGI/ASI will cure cancer, help us colonize Mars, and make lattes by 2026.

Meanwhile, Sam Altman’s over here saying, “Guys, chill we haven’t even found AGI yet,” while ChatGPT still hallucinates ways to accidentally DDoS websites. (Pro tip: Don’t ask it to “crawl similar URLs” unless you want chaos.)

But then… China dropped a mic.

DeepSeek R1: Open-Source Brains, No Subscription Fees

Meet DeepSeek R1, new open-source AI model that’s basically screaming, “Why pay OpenAI $200/month when you can have this for free?” It’s designed for Chain of Thought reasoning - meaning it solves problems step-by-step, like a human scribbling notes in a math class.

Why it’s cool:

• Reinforcement learning magic: Instead of spoon-feeding it answers, R1 figures things out on its own. Think of it as a digital toddler who learns to walk by face-planting 1,000 times.
• Size matters (kinda): The full version has 671 billion parameters (translation: big brain energy), but even the “mini” 32B version rivals OpenAI’s smaller models.
• Transparent thinking: Ask it to solve a math problem, and it’ll show its work like a nerdy show-and-tell.

But Wait… Should We Trust Benchmarks?

DeepSeek claims R1 beats OpenAI in math and coding tasks. But and this is a big “but” benchmarks can be… sketchy. Remember when we found out some benchmark companies are funded by OpenAI? ??

My take? Vibes > numbers. Try the model yourself. Does it feel smarter? Does it write better code than you? That’s the real test.

When LOKI is an security engineer

While working on my hypothesis of having an super cool (will name it later) defence framework... ended up running a fun experiment.

The problem: Hackers scan for open ports (like digital doorways) to find vulnerabilities. Fewer open ports = fewer attack options. But what if all ports looked open?

The solution: Techniques like port spoofing create a smokescreen. It makes all 65,535 ports appear open, drowning attackers in fake targets. Imagine handing a burglar 65,000 keys—they’ll waste hours trying locks that lead nowhere.

How it works:

• Redirects port scans to a decoy service
• Sends fake "banners" (e.g., “This is totally SSH, trust me!”) to mislead attackers.
• Logs their attempts, letting you spot threats early

Why it’s fun: You can even customize fake responses. Port 21? Send “gotcha” Port 80? “404 Error: Try harder.” ??

But...

• Requires setup (iptables, config files)
• Best for labs or honeypots - don’t annoy real users
• I recommend using it in a CTF competition, will be an absolute fun

Takeaway: Cyber defense isn’t just firewalls. Sometimes, it’s pranking bad actors into rage-quitting.

New Moore’s Law - The number of people predicting the death of Moore's law doubles every two years

From when I picked up the technology, moore's law has died like 900 times by now ??. Lightmatter 's work is interesting, achieving the speed of 100 terabits/sec - no electrical bottlenecks but somewhere inside I feel this is NOT computing, it's just faster BUS speed. From what I know light does NOT travel at the speed of light through optical fiber. It travels at v = c/n which is speed of light, c, divided by the refractive index, n, of the optical fibre core. Light only travels at speed of light in a vacuum. At 10:45 timestamp "Imagine data flowing through the computer at literally the speed of light... while traditional computers are stuck waiting for electrical signals to bounce back and forth... light moves through these chips without any delay." There is a lot of misleading things here but I am sure guys from lightmatter knows well on what they are working on. For who I am, the questions pops up in my mind are

1. What actual delays in existing architectures can be solved with this?
2. What is the current speed and what do you expect the speed to be with the photonic system?
3. How much performance does this get you on the chip?

Maybe I will spend some time on these questions(depends). Anyway below is the video for you guys to understand it by yourself.

The Secret Life of a Linux Packet (It’s a Jungle in There)

The TL;DR:

1. Your app (like Chrome) shouts, “I NEED DATA!”
2. The kernel grabs a SK buff (a fancy data bucket) and stuffs your request inside.
3. TCP/UDP adds headers like shipping labels (“Fragile Don’t drop me!”).
4. IP layer plays GPS, routing packets through the digital wilderness.
5. Network card yeets it onto the internet highway.

Cool tidbits:

• SK buffs are ninjas: They share data between processes without copying - just pointers. Efficient!
• TCP vs. UDP: TCP is that friend who texts “Did you get my message???” UDP is the “YOLO, send and forget” type.
• Everything’s a file: Sockets? Just file descriptors in disguise. Unix philosophy, baby.

Why care?

• Optimization: Knowing this helps squeeze speed from servers.
• Debugging: When packets vanish, you’ll know where to hunt.

Vibe check: Linux networking is like a Rube Goldberg machine - chaotic, but it works.

Check out the paper...

Phew, a lot has happened this week, but I didn’t want to info-dump here, so I have shared just a few handpicked highlights. thanks for sticking with me! I hope you found something valuable, whether it was a new insight, a fresh perspective, or just a relatable struggle. If not, that’s okay too sometimes it’s about enjoying the process and exploring ideas together also feel free to share your thoughts, feedback, or even your own experiences - I had love to hear from you.

- Prince B