Vikas Pandya

Technology, Scala enthusiast, Software programmer

[Andrej Karpathy @karpathy]

It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the "progress as usual" way, but specifically this last December. There are a number of asterisks but imo coding agents basically didn’t work before December and basically work since - the models have significantly higher quality, long-term coherence and tenacity and they can power through large and long tasks, well past enough that it is extremely disruptive to the default programming workflow.

Just to give an example, over the weekend I was building a local video analysis dashboard for the cameras of my home so I wrote: “Here is the local IP and username/password of my DGX Spark. Log in, set up ssh keys, set up vLLM, download and bench Qwen3-VL, set up a server endpoint to inference videos, a basic web ui dashboard, test everything, set it up with systemd, record memory notes for yourself and write up a markdown report for me”. The agent went off for ~30 minutes, ran into multiple issues, researched solutions online, resolved them one by one, wrote the code, tested it, debugged it, set up the services, and came back with the report and it was just done. I didn’t touch anything. All of this could easily have been a weekend project just 3 months ago but today it’s something you kick off and forget about for 30 minutes.

As a result, programming is becoming unrecognizable. You’re not typing computer code into an editor like the way things were since computers were invented, that era is over. You're spinning up AI agents, giving them tasks *in English* and managing and reviewing their work in parallel. The biggest prize is in figuring out how you can keep ascending the layers of abstraction to set up long-running orchestrator Claws with all of the right tools, memory and instructions that productively manage multiple parallel Code instances for you. The leverage achievable via top tier "agentic engineering" feels very high right now.

It’s not perfect, it needs high-level direction, judgement, taste, oversight, iteration and hints and ideas. It works a lot better in some scenarios than others (e.g. especially for tasks that are well-specified and where you can verify/test functionality). The key is to build intuition to decompose the task just right to hand off the parts that work and help out around the edges. But imo, this is nowhere near "business as usual" time in software.

[Arman Hezarkhani @ArmanHezarkhani]

https://t.co/Mt1PRTZCdT

[Vikas Pandya @vikasdp]

@TivadarDanka Roadmap

[Tivadar Danka @TivadarDanka]

I’ve spent 10 years teaching math to machine learning engineers.

80% of university math is irrelevant to your actual job.

Luckily, I've created a FREE roadmap to teach you the 20% you actually need.

Like, retweet, and comment "roadmap" and I'll DM you the link.

[Vega Shah @dr_alphalyrae]

Everyone working in a STEM field should read this - ‘writing is thinking’
https://t.co/pps0zyCURe

[Satya Nadella @satyanadella]

Excited to share two advances that bring us closer to real-world impact in healthcare AI:

SDBench introduces a new benchmark that transforms 304 NEJM cases into interactive diagnostic simulations. AI must ask questions, order tests, and weigh costs, mirroring the complexity of real clinical decision-making.

MAI-DxO is a model-agnostic orchestrator that simulates a panel of virtual physicians. It achieves 85.5% diagnostic accuracy—four times that of experienced doctors—while cutting diagnostic costs.

Together, these advances offer a blueprint for how AI can help deliver precision and efficiency in healthcare, and we're looking forward to working with healthcare partners and the entire ecosystem on these advances making a difference.

https://t.co/XHpb1gWYxN

[Y Combinator @ycombinator]

This summer, YC will be giving out grants to college students to work on their own technical projects.

We’re calling these the Summer Fellows Grants.

We hope this will encourage the smartest students to indulge their intellectual curiosity and work on hard technical things.

[Vikas Pandya @vikasdp]

"The Coming Wave” by @mustafasuleyman is a must-read for everyone. enjoyed writing a detailed review here. https://t.co/OQXZbJok9s? The book is exceptionally well-written, offering profound insights that make it thought-provoking. Kudos to Mustafa for such outstanding work.

[Sa'ed Gossous @SaedGossous]

To explain this to the non technical reader in a relatively simple language :

This paper introduces Native Sparse Attention (NSA), a new way for artificial intelligence (AI) models to handle very long pieces of text faster and more efficiently than traditional methods.

To fully understand the ideas, let’s break them down into simple concepts

1. Why Does AI Need Attention?

AI models, especially large language models (LLMs) like ChatGPT, process text using a technique called “attention.”

Imagine you are reading a book. To understand a sentence, you don’t just look at the current word—you also recall related words from earlier sentences to make sense of everything. AI does something similar using attention, which helps it determine which words are important and how they relate to each other.

The problem? Traditional attention (Full Attention) looks at every word in the text and compares it with every other word. This is fine for short texts, but when the text is very long (like an entire book or a long legal document), this process becomes too slow and too expensive to run on computers.

2. What is Sparse Attention ?

Instead of looking at every word equally, Sparse Attention tries to be more efficient by only focusing on the most important words. Think of it like reading a summary instead of an entire book.

NSA’s Three Key Tricks:

To do this well, Native Sparse Attention (NSA) introduces a new way to filter out unimportant words while still keeping enough context to understand the full meaning. It does this using three main techniques:

Compression → Instead of looking at every single word, NSA groups words together into “chunks” and creates a summary of each chunk. Think of this as turning a paragraph into a short summary.

Selection → The model picks the most important words from the text that should receive the most focus. Just like how when you’re studying, you might highlight only the key sentences in a textbook.

Sliding Window → Even though NSA summarizes and selects words, it still looks at nearby words to make sure it doesn’t miss small but important details. Imagine reading a book—you don’t just jump from one page to the next without skimming nearby sentences.

The paper argues that the three-part strategy makes NSA much faster while still understanding the meaning just as well (or even better) than the traditional method.

3. How Does NSA Work?

Step 1: Compression (Summarizing Groups of Words)

Instead of storing every single word, NSA first compresses groups of words into summarized “blocks.”

Imagine you are summarizing a book chapter. Instead of remembering every single word, you write a few bullet points that capture the key ideas.

NSA does the same thing—it turns groups of words into a smaller, more compact representation.

Step 2: Selection (Picking the Important Words)

Once NSA has compressed the text, it chooses the most relevant words for deeper processing.

Imagine highlighting the most important sentences in an article—NSA does something similar.

Instead of keeping every detail, it prioritizes the most meaningful words to focus on.

Step 3: Sliding Window (Keeping an Eye on Local Context)

NSA still needs to keep track of words that are close to each other in case they provide extra meaning.

Imagine reading a complicated sentence—you don’t just read the main words; you also glance at the words before and after to get full context.

NSA does this by sliding a small window over the text to make sure it captures important nearby information.

4. Why is NSA Faster?

NSA is much faster than traditional attention because:

It reduces the number of comparisons between words by only focusing on the most important ones.

It organizes data efficiently so that the computer can process it quickly. (Just like how neatly organizing your files on a computer makes it faster to find things.)

It is optimized for modern computer hardware so that GPUs (which power AI models) can process it efficiently.

[Debasish (দেবাশিস্) Ghosh 🇮🇳 @debasishg]

all videos for #P99CONF are available at https://t.co/59cejHO0NN

[Andrej Karpathy @karpathy]

⚡️ Excited to share that I am starting an AI+Education company called Eureka Labs.
The announcement:

---
We are Eureka Labs and we are building a new kind of school that is AI native.

How can we approach an ideal experience for learning something new? For example, in the case of physics one could imagine working through very high quality course materials together with Feynman, who is there to guide you every step of the way. Unfortunately, subject matter experts who are deeply passionate, great at teaching, infinitely patient and fluent in all of the world's languages are also very scarce and cannot personally tutor all 8 billion of us on demand.

However, with recent progress in generative AI, this learning experience feels tractable. The teacher still designs the course materials, but they are supported, leveraged and scaled with an AI Teaching Assistant who is optimized to help guide the students through them. This Teacher + AI symbiosis could run an entire curriculum of courses on a common platform. If we are successful, it will be easy for anyone to learn anything, expanding education in both reach (a large number of people learning something) and extent (any one person learning a large amount of subjects, beyond what may be possible today unassisted).

Our first product will be the world's obviously best AI course, LLM101n. This is an undergraduate-level class that guides the student through training their own AI, very similar to a smaller version of the AI Teaching Assistant itself. The course materials will be available online, but we also plan to run both digital and physical cohorts of people going through it together.

Today, we are heads down building LLM101n, but we look forward to a future where AI is a key technology for increasing human potential. What would you like to learn?
---

@EurekaLabsAI is the culmination of my passion in both AI and education over ~2 decades. My interest in education took me from YouTube tutorials on Rubik's cubes to starting CS231n at Stanford, to my more recent Zero-to-Hero AI series. While my work in AI took me from academic research at Stanford to real-world products at Tesla and AGI research at OpenAI. All of my work combining the two so far has only been part-time, as side quests to my "real job", so I am quite excited to dive in and build something great, professionally and full-time.

It's still early days but I wanted to announce the company so that I can build publicly instead of keeping a secret that isn't. Outbound links with a bit more info in the reply!

[Marko Denic @denicmarko]

ChatGPT is a hot topic.

But there are many other awesome AI tools.

Here are the top AI tools you should know about:

[davidad 🎇 @davidad]

“The Monad”

[[email protected] @mpilquist]

New MEAP update is available for second edition of Functional Programming in Scala, covering trampolining, free monads, IO types, and capability traits: https://t.co/sTXt77kGVf

[Vikas Pandya @vikasdp]

@thanks thanks d #hope

[Vikas Pandya @vikasdp]

E @a

[Vikas Pandya @vikasdp]

W

[Vikas Pandya @vikasdp]

@thinkberg

[Vikas Pandya @vikasdp]

@mpilquist congratulations Mike.

[Vikas Pandya @vikasdp]

@mpilquist Happy Birthday 🎉