Nikhil

Tinker!

[Nikhil @nik_balaji]

Jev + Graphical Model -> back to probabilistic graphical models 🤔

[Nikhil @nik_balaji]

An Interesting new RL paper (KLPO): it aims to provide the stable updates of traditional trust-region or KL-regularized methods without requiring computationally heavy value networks or multi-sequence response grouping. https://t.co/O7IhCxuV2v

[François Chollet @fchollet]

Sufficiently advanced agentic coding is essentially machine learning: the engineer sets up the optimization goal as well as some constraints on the search space (the spec and its tests), then an optimization process (coding agents) iterates until the goal is reached.

The result is a blackbox model (the generated codebase): an artifact that performs the task, that you deploy without ever inspecting its internal logic, just as we ignore individual weights in a neural network.

This implies that all classic issues encountered in ML will soon become problems for agentic coding: overfitting to the spec, Clever Hans shortcuts that don't generalize outside the tests, data leakage, concept drift, etc.

I would also ask: what will be the Keras of agentic coding? What will be the optimal set of high-level abstractions that allow humans to steer codebase 'training' with minimal cognitive overhead?

[dax @thdxr]

ai is such a magnifier of judgement that it's going to get really obvious who knows what they're doing and who doesn't

[Gurwinder @G_S_Bhogal]

“Nothing gives you a clearer look into someone than how they misinterpret things; every misinterpretation is a confession.”

—@DylanoA4

[Mihail Eric @mihail_eric]

All assignments for Stanford's The Modern Software Developer are now available online.

This is the first comprehensive university course covering how coding LLMs are transforming every stage of the software development life cycle. The assignments are intended to take you from noob to expert in how to use AI to improve your software engineering productivity.

Enjoy!

https://t.co/YxzneOSCY5

[Nikhil @nik_balaji]

Skills based prompting setup.
https://t.co/A34p8H1qB8

[Andrej Karpathy @karpathy]

I was inspired by this so I wanted to see if Claude Code can get into my Lutron home automation system.

- it found my Lutron controllers on the local wifi network
- checked for open ports, connected, got some metadata and identified the devices and their firmware
- searched the internet, found the pdf for my system
- instructed me on what button to press to pair and get the certificates
- it connected to the system and found all the home devices (lights, shades, HVAC temperature control, motion sensors etc.)
- it turned on and off my kitchen lights to check that things are working (lol!)

I am now vibe coding the home automation master command center, the potential is 🔥.And I'm throwing away the crappy, janky, slow Lutron iOS app I've been using so far. Insanely fun :D :D

[elvis @omarsar0]

@TedHZhang Yes. It starts in two weeks: https://t.co/lmI57QlqVs

[Eren Bali @erenbali]

Yes. Software engineers will all go away. There will be a new profession of people whose only job is to tell the machines what they should do in a manner that’s efficient and precise.

[Nikhil @nik_balaji]

@heyanuja @papertrailshq Looks great!

[Nikhil @nik_balaji]

Thought for the day.

[Ed Latimore @EdLatimore]

Embarrassment is the cost of entry.

If you aren't willing to look like a foolish beginner, you'll never become a graceful master.

[Nikhil @nik_balaji]

Loving #Obsidian so far, especially the 'add links to local pdfs' feature. To reference a specific page within a PDF in Obsidian, use the syntax [[filename.pdf#page=number]]

[Nikhil @nik_balaji]

Trying out #Obsidian today. Lets see how it goes! #Evernote is good but I'm intrigued by the idea of building a second brain.
PS: Graph View and Linked Mentions seem like something I'd be very interested in.

[Nikhil @nik_balaji]

Better than general purpose FMs when:
* Physical/Domain-Specific laws need to be respected
* Data is scarce
* Generalization beyond training data is essential

[Nikhil @nik_balaji]

Towards Physics-Guided Foundation Models(FMs): new approach of building domain-specific FMs for scientific/engineering tasks. Don't rely on big data/ gen-purpose architectures alone, embed domain knowledge into the model arch, objectives, training data.
https://t.co/p3d20AiYtw

[Nikhil @nik_balaji]

@oh_that_hat frontier names

[Nikhil @nik_balaji]

@neural_avb Using transfer learning to train models and acutally get good working results. https://t.co/p8vtHip1hC

[Jeremy Howard @jeremyphoward]

“Fun with small image data-sets” by @nikhilbalaji https://t.co/Kjpaxi0MOG
Shows how to download <50 images from Google and use them to create a near perfect image classifier!