Suban Shrestha

[AI Safety Nepal @aisafetynepal]

Are you using AI to think more; or think less?

At our kick-off event on May 17, @baibhavbista , Member of Technical Staff at Roam Research, will be talking about cognitive delegation vs surrender.

⌛Deadline: May 10
Register: https://t.co/TqzbK5piKa

[sads @_saddicted]

Prime minister of nepal @kpsharmaoli is killing students in uniform during a peaceful protest! @UNICEF @UN @OHCHRAsia @hrw @BBCWorld

[Suban Shrestha @zetoxtz]

@suparth_ghimire Hmmm

[Тsфdiиg @tsoding]

The long term value of Recreational Programming (apart from enabling you to enjoy what you are doing again) comes from allowing you to explore unusual things outside of the local optimum that the market pressure forces you into.

[Suban Shrestha @zetoxtz]

Really solid insights on AI agent context engineering here.

Faced some weird hallucinations on a recent project - we were accidentally "few shotting" ourselves with poor context management.

Cool to see this pattern getting a proper name.

[Yichao 'Peak' Ji @peakji]

After four overhauls and millions of real-world sessions, here are the lessons we learned about context engineering for AI agents: https://t.co/Ql014rEzBQ

[Suban Shrestha @zetoxtz]

@39abii_ Lets gooo 🥳🥳

[Suban Shrestha @zetoxtz]

#LSPPDay13

Read the protocol docs for python.
A cool feature similar to the Interfaces in go.
Does help skip inheritance.

I use case could be for typing classes you have less control over.

#60DaysOfLearning2025 #LearningWithLeapfrog
@lftechnology

[Suban Shrestha @zetoxtz]

#LSPPDay12

Cool talk about virtual environments in python.
https://t.co/JYLowCMLEN

#60DaysOfLearning2025 #LearningWithLeapfrog
@lftechnology

[Suban Shrestha @zetoxtz]

#LSPPDay11

More on maps today. Went though the common data types used. And skimmed though the geojson RFC.

https://t.co/li9zBOMLam

Surprisingly not that hard to read.

#60DaysOfLearning2025 #LearningWithLeapfrog
@lftechnology

[Suban Shrestha @zetoxtz]

#LSPPDay10
- Had to look at maps today

re learned working with postgis, data transformations, and queries.

#60DaysOfLearning2025 #LearningWithLeapfrog
@lftechnology

[Suban Shrestha @zetoxtz]

#LSPPDay09
- Khan academy's financial literacy

Apparently, Big expenses won't hit you like a train wreck if you plan ahead by saving a set amount each month for expenses you know are coming.

#60DaysOfLearning2025 #LearningWithLeapfrog @lftechnology

[Suban Shrestha @zetoxtz]

It also suggests using tool call as a way to contact humans , which is something i want to experiment with but have never done.

[Suban Shrestha @zetoxtz]

#LSPPDay08
light day today
read part of 12-factor-agents: https://t.co/3hhW6glzOg
It talks about owning the prompt, context and control flow rather than handing to to a library, as it gives us more room for experiments.

#60DaysOfLearning2025 #LearningWithLeapfrog
@lftechnology

[Suban Shrestha @zetoxtz]

@lftechnology a small write up on this:
https://t.co/s18Exy0i8f

[Suban Shrestha @zetoxtz]

#LSPPDay07

Read a paper from Apple demonstrating complete accuracy collapse of LLM/LRM's reasoning capability once problems become complex enough.

For this, they used the famous Tower of Hanoi problem. That reminds me
#60DaysOfLearning2025 #LearningWithLeapfrog @lftechnology

[Suban Shrestha @zetoxtz]

#LSPPDay06
Talks on LLM Evals
One intriguing analogy compares LLM outputs to basketball shots on a court.
Showcasing
- inherently unreliable - even pros miss shots
- Making sure we cover the important parts of the Court

#60DaysOfLearning2025 #LearningWithLeapfrog
@lftechnology

[Suban Shrestha @zetoxtz]

A small note one of the talks:
Instacart's LLM Driven Search and Discovery:
https://t.co/2tzuc2NVHN

[Suban Shrestha @zetoxtz]

#LSPPDay05
Spent time watching:
the talks from AI Engineer World’s Fair 2025
https://t.co/r2FUmh2Il1

The LLM RECSYS track has some cool usages of AI to enhance recommendation systems.
#60DaysOfLearning2025 #LearningWithLeapfrog
@lftechnology

[Suban Shrestha @zetoxtz]

#LSPPDay04
Read more on Evals for LLM products today.

This be the mental model for improving AI systems. according to Hamel Husain

My hands are itching to build this at work.
@lftechnology
#60DaysOfLearning2025 #LearningWithLeapfrog

[Suban Shrestha @zetoxtz]

@asminsilwal_ @lftechnology I write today, please forgive 😭

[Suban Shrestha @zetoxtz]

#LSPPDay03

Read through Hamel Husain's recent blogs on LLM Evals. (LLM Eval FAQ and A Field Guide to Rapidly Improving AI Products)
Things are still quite fuzzy for me, will try implementation tomorrow.

#60DaysOfLearning2025 #LearningWithLeapfrog
@lftechnology