This Week I Learned - Week 38 2025

This Week I Learned - 

* "Software testing methodologies such as Test Driven Development (TDD), a test-intensive approach that involves first writing rigorous tests for correctness and only then making progress by writing code that passes those tests, are an important way to find bugs. But it can be a lot of work to write tests. (I personally never adopted TDD for that reason.) Because AI is quite good at writing tests, agentic testing enjoys growing attention." - Andew Ng

* "An LLM agent runs tools in a loop to achieve a goal." - Simon Willison

* Free Hands-On Microsoft Azure Labs 

* Xipu Li on the future of programming -  If you're learning to code today, please don't stop. But don't make coding your only skill. The developers who thrive in this new world will be the ones who understand users, markets, and business models as well as they understand technology.

If you're already a developer or building a company, focus on:

• Better user understanding. 
• Stronger distribution channels. 
• Clearer market positioning. 
• Faster learning cycles. 
• Better taste in what to build. 
• Superior execution on the parts that can't be automated

Three things become incredibly valuable when technical implementation gets commoditized -

• What people actually need.
• Knowing what to build and what not to build. 
• Getting products in front of the right people and convincing them to care.

The code itself has become an implementation detail, like the electrical wiring behind your drywall. You know it's there; you trust that it works, but you don't think about it unless something breaks. 

Developers will have to start spending most of their time doing what product managers traditionally do - understanding requirements, translating them into clear specifications, writing & reviewing design documents. 

* As in Twitter, Grok AI can be called upon in any Reddit thread, by mentioning u/AskGrok. As per Grok, it is "powered by xAI, Elon Musk's venture, which is bankrolled by a mix of investors (including Musk himself) and likely offset by premium subscriptions. No shadowy overlords, just good old-fashioned venture funding and user upgrades keeping the lights on."

Its presence on Reddit via the AskGrok account isn't powered by a direct partnership between xAI (or X, formerly Twitter) and Reddit. Instead, it leverages publicly available Reddit APIs and third-party tools like Zapier for integrations, allowing it to pop into threads when mentioned. No secret handshakes or exclusive deals—it's all about open access, though xAI handles the backend compute. 

Key venture capitalists and investors include Andreessen Horowitz, Lightspeed Venture Partners, Sequoia Capital, Tribe Capital, Fidelity, BlackRock, TPG, and others who've chipped in across rounds totaling over $12 billion since 2024.

* "...all text generated from an LLM is hallucinated. The generations just happen to be correct most of the time." - OCI Generative AI Professional course

* For more than a decade, consumers have fumbled with language translation apps on their phones that were awkward to use, like Google Translate and Microsoft Translator. They required users to hold their phone’s microphone up to a person speaking a foreign language and wait for a translation to be shown on a screen or played through the phone’s tiny speakers. The translations were often inaccurate.

AirPods Pro 3 has up to 4x more Active Noise Cancellation than AirPods Pro and AirPods 4 with ANC

In contrast, AirPods users need only to make a gesture to activate the digital interpreter. About a second after someone speaks, the translation is played in the wearer’s preferred language through the earbuds. 

The AirPods’ reliance on large language models, the technology that uses complex statistics to guess what words belong together, should make translations more accurate than past technologies - NYT

* Gall's Law is a rule of thumb for systems design. It states:

A complex system that works is invariably found to have evolved from a simple system that worked. A complex system designed from scratch never works and cannot be patched up to make it work. You have to start over with a working simple system.

This law is essentially an argument in favour of underspecification: it can be used to explain the success of systems like the World Wide Web and Blogosphere, which grew from simple to complex systems incrementally, and the failure of systems like CORBA, which began with complex specifications. Even massive systems like Netflix, Google, and Facebook began on a small scale and grew step by step over decades. Gall's Law has strong affinities to the practice of agile software development. This philosophy can also be attributed to extreme programming, which encourages doing the simplest thing first and adding features later.

* Microsoft has roughly 5,200 employees with H1-B visas.

* BharatGen, an IIT Bombay-led startup is building a one trillion-parameter LLM, which will be distilled into smaller systems for specific sectors. 

* Fractal Analytics is developing India's first large reasoning model, a 70 billion-parameter system focused on healthcare and STEM.

* The government is exploring a shift in how it evaluates state-run companies. It is considering technology adoption and artificial intelligence (AI) deployment as new benchmarks for PSU classification under the Maharatna-Navratna-Miniratna framework, moving beyond traditional financial metrics such as net worth, turnover, and profitability.

* India is making big investments in chip manufacturing: Tata and PSMC are investing about ₹91,000 crore in 28nm chip plants, Micron is investing ₹22,500 crore on other mature chip technologies, and Vedanta-Foxconn planned a ₹1.54 lakh crore investment but that project is currently stalled. 

VIKRAM is India’s first chip made with 28-nanometer (nm) technology, announced in September 2025.

The 28nm chip technology has been around globally since about 2011 to 2014. For comparison, the most advanced chips today use 3nm technology, which started around 2022.

Globally, around 70-90% of all chips made and sold use mature (older) technologies like 28nm or higher, not the latest cutting-edge.

India imports chips worth about $50-65 billion every year, mostly the mature, well-established types.

Building chip factories for mature technology chips costs less (about $3-5 billion) compared to the very advanced ones (like $20-30 billion for 3nm), and mature fabs produce more usable chips per batch (80-90% yield vs 60-80%).

Companies like TSMC actually make more profit from manufacturing these mature node chips than from the newest technology chips.

India is focusing on building solid, cost-effective chip manufacturing with proven technology while making strides to reduce chip imports and strengthen the semiconductor industry.

* Cosmic Orange has become one of the most popular color choices for the iPhone 17 in India. - MoneyControl

* SIM boxes enable cyber fraudsters to convert international calls into local ones, making it easier to deceive victims and evade law enforcement. Here's how it works: A fraudster uses a VoIP dialer app to make a call, which lands on a high-speed broadband IP address in India, managed by the cyber fraud gang's Indian associate. The broadband connects the call to a SIM box device, functioning like a mobile handset with a specific IMEI number. This device links the VoIP call to the intended Indian phone number using multiple SIM cards. Each SIM box can hold 100 to 500 SIM cards, enabling several fraudsters to target different victims at once. These gangs often use SIM boxes for digital arrest scams, where victims see a local number and are tricked into believing the caller is a police or CBI officer. Their usage has surged since the launch of the International Incoming Spoofed Call Prevention System (IISCPS) last year. In October 2024, the DoT introduced this system to detect fraudulent international calls posing as Indian numbers. IISCPS identified and blocked 1.35 crore spoof calls within 24 hours.

* CBFC Watch is an open-source, searchable archive of every movie certified by the censor board of India since 2017 (before the portal was modified in a way that made automated collection impossible in June of this year). That is almost 18,000 movies and over 100,000 modification details made machine-analyzable. You can search for specific movies, terms, actors, directors, languages, categories of censorship, or really whatever you can think of. Each film page comes with an analysis of how that movie compares to others like it. 

Film certifications fall into broadly three groups — adults only (A), unrestricted for all ages (U), and parental guidance/general audience (UA). Did you know, the same film, when released in different languages, can received different ratings.

* Research shows that people with larger necks relative to their body size face increased risks of several serious health conditions. Essentially, neck circumference serves as a proxy for visceral fat — the harmful fat that wraps around your organs

* A whales' perspective to see how they hunt in the deep sea.