Vansh soni

Building things to understand how they work

[Vansh soni @_VanshSoni_]

Jev is free to use till sept 25

[Vercel Developers @vercel_dev]

Jev by @typesafeai is free on Vercel AI Gateway until Sept 25.

Build with the fastest adopted model on the Gateway at no cost.

https://t.co/KsFHUhmaLe
https://t.co/7KsKeZElra

[Vansh soni @_VanshSoni_]

Cracks showing. OAI & Anthropic publicly seek AI R&D slowdown.

Trans: tech no more useful now, growth stalled. 1st to cut $ burn faces mkt wrath. Claim bogeyman tech ends Earth life so all slow together. Lobby impose via safety/reg. Then US gov bails. Yay!

If NVDA=bank, start sweating. Rare for firms to ask voluntary slowdown.

[Vansh soni @_VanshSoni_]

if you're confused about Jev this is 100% worth your time

[Sydney Runkle @sydneyrunkle]

https://t.co/CF54CU9Ddz

[Vansh soni @_VanshSoni_]

A new cybersecurity investigation reveals LG Smart TVs are quietly spying on living rooms: recording audio even when screens look off, tracking nearby phones on your home Wi-Fi, and cataloging everything on-screen for ad brokers. Security researchers urge owners to disconnect them from the internet immediately

https://t.co/ZPRopNp0Y0

[Vansh soni @_VanshSoni_]

A new cybersecurity investigation reveals LG Smart TVs are quietly spying on living rooms: recording audio even when screens look off, tracking nearby phones on your home Wi-Fi, and cataloging everything on-screen for ad brokers. Security researchers urge owners to disconnect them from the internet immediately

https://t.co/ZPRopNp0Y0

[Vansh soni @_VanshSoni_]

Chinnaswamy stadium at night just hits different. This is how the ground looks under the lights.

[Vansh soni @_VanshSoni_]

I have written an article about complete backend how things are connected, stored, processed. Check it out

[Vansh soni @_VanshSoni_]

Most backend engineers know the terms.

But do you really understand how they connect?

Code → Memory → CPU → OS → Concurrency → I/O → Networking → Storage → Databases → Distributed Systems → Reliability → Performance. https://t.co/1gbMC8i7RU

[Vansh soni @_VanshSoni_]

Most backend engineers know the terms.

But do you really understand how they connect?

Code → Memory → CPU → OS → Concurrency → I/O → Networking → Storage → Databases → Distributed Systems → Reliability → Performance. https://t.co/1gbMC8i7RU

[Vansh soni @_VanshSoni_]

HSR, 5:30 AM. Zero honks, just vibes.

[Vansh soni @_VanshSoni_]

HSR, 5:30 AM. Zero honks, just vibes.

[Vansh soni @_VanshSoni_]

If you are preparing for System Design Interviews, This one is simply THE BEST. Get everything at one place. GitHub: https://t.co/P2YZ7T7fZ7

[Vansh soni @_VanshSoni_]

3 API mistakes that can seriously hurt your production system:

1. Skipping Authentication & Authorization
2. Not using Pagination for large datasets
3. Exposing detailed backend errors to clients

They look like small implementation decisions.

But they can turn into:
→ Data leaks
→ Performance issues
→ Security vulnerabilities

Fix these before you ship, not after production breaks.

Save this for your next API review. 🔐

[Vansh soni @_VanshSoni_]

Everyone is talking about GPT-6 Astra’s 99.9% score, but here is the precise context.
ARC Prize ran and published both evaluations. Astra scored 62.7% with the Standard harness at max reasoning, costing $26,098. It scored 99.9% with the Provider Adapter at high reasoning, costing $18,817. The Adapter preserves reasoning between calls, so the difference came from how context was carried forward, not who ran the test.
Claude Opus 5 scored 30.2% at high reasoning with the Standard harness. Therefore, 62.7% versus 30.2% uses the same harness, but different reasoning settings.
More importantly, ARC-AGI-3 tests whether AI can explore a completely new game, learn its rules and complete the goal. It doesn’t prove AGI is here, but it shows how quickly AI is improving.
Don’t wait for AGI. Start using AI for one repeated task today.

[Kush Bang @KushBang1]

We need an uber silent in bangalore
@rahil_mavani @_VanshSoni_ @Uber

[Elon Musk @elonmusk]

@chesscom Wrong, all legal chess positions are ~40 oom less than the estimated atoms in the Universe. And ASI will figure out ways to solve and compress that are far beyond what we could possibly comprehend.

(Here I am arguing with some random intern at a chess website 🤦‍♂️)

[Vansh soni @_VanshSoni_]

Tired of burning tokens on AI coding? GitHub just broke down why blindly shortening outputs actually drives costs up, showing that real savings come from cleaning up hidden junk like build logs and eliminating repeat retrieval steps
https://t.co/PsKF5OjBPk

[Vansh soni @_VanshSoni_]

What if we build a lightweight, open-source Google Authenticator in Rust from scratch, who'd use the CLI?

[Vansh soni @_VanshSoni_]

@shydev69 That’s what it’s like 5 min of topology creation

[Vansh soni @_VanshSoni_]

@Arnav_Upadhyay7 It’s easy u just have to give ip addresses, gateway route, routing policy and acl if needed isn’t it like less than 10 min of topology?