_arnav_

ML, DL, AI, LLM || Experimenting+Building

[_arnav_ @arnava_]

Getting ghosted after the final interview hits different. 🫠

Made it to the Final Round at @arya_ai1.

Discussed:
• Salary expectations
• Joining date
• Projects
• Company culture

Then… silence.

Tried reaching out. No response.

After multiple rounds, candidates deserve a clear yes or no.

Has this happened to you?

[_arnav_ @arnava_]

@MrrrOzi @arya_ai1 True that🤣
But ig very small changes cumulatively shape the industry overall

[_arnav_ @arnava_]

Round 2 at @arya_ai1!

This time, I spoke with the Head of Research.

We went deep into:
• Transformer architecture
• The current state of Transformers
• Where the field could be heading next
• My views on the future of foundation models

Follow for Round 3!

[_arnav_ @arnava_]

@ishita_here_ Will do!
Thanks!

[_arnav_ @arnava_]

Why can’t we just give text directly to an LLM?

Characters → Unicode → bytes → tokens → subwords.

A first-principles look at the problems each representation solves, and why we eventually arrive at subword tokenization. https://t.co/seTfOnt52h

[_arnav_ @arnava_]

@Kpaxs Can't reddit be a place for that, where people learn from each other's experiences?
Will have to filter through a lot though😅

[_arnav_ @arnava_]

I applied to @arya_ai1 for a Research Scientist role a few months back. Sharing the interview questions I faced to help others preparing for similar AI/ML roles.

Here’s what was asked in Round 1:
• Resume-based in-depth technical questions
• 2 Sum problem
• Design a data structure supporting add, present, edit, and prefix search

Comment below if you want to know what was asked in Round 2 👀

[_arnav_ @arnava_]

@tanejagarvit_ Oh, understood

[_arnav_ @arnava_]

@SPA59533577 @indiainpixels Thanks for sharing!

[_arnav_ @arnava_]

@ishita_here_ Thanks @ishita_here_ !
Any next topic you wanna explore?

[_arnav_ @arnava_]

https://t.co/2GO3WJG0Oz

[_arnav_ @arnava_]

Why can’t we just give text directly to an LLM?

Characters → Unicode → bytes → tokens → subwords.

A first-principles look at the problems each representation solves, and why we eventually arrive at subword tokenization. https://t.co/seTfOnt52h

[_arnav_ @arnava_]

Wanna follow @paulg's advice and build your own LLM?
Start here!

[Paul Graham @paulg]

Someone asked what I'd do if I were 17. I'd learn how to build LLMs from scratch, and then train ones as powerful as I could with whatever hardware I could get access to.

[_arnav_ @arnava_]

https://t.co/2GO3WJG0Oz

[_arnav_ @arnava_]

Why can’t we just give text directly to an LLM?

Characters → Unicode → bytes → tokens → subwords.

A first-principles look at the problems each representation solves, and why we eventually arrive at subword tokenization. https://t.co/seTfOnt52h

[_arnav_ @arnava_]

This is the best article about how AI breaks text into pieces!
Thank me later!

[_arnav_ @arnava_]

Why can’t we just give text directly to an LLM?

Characters → Unicode → bytes → tokens → subwords.

A first-principles look at the problems each representation solves, and why we eventually arrive at subword tokenization. https://t.co/seTfOnt52h

[_arnav_ @arnava_]

Hoping to meet incredible people in Hyd on 18 Sept!

[Aakrit Vaish @aakrit]

We’re taking Mixture of Experts across India 🇮🇳

Something I’ve wanted to do since we started Activate — get on the road to meet the best AI founders, researchers and builders wherever they are.

6 cities. 5 weeks. Starting next week.

Here we go 🧵👇

[_arnav_ @arnava_]

@Hi_Mrinal @mohitwt_ @aryagxr @jino_rohit @sodakeyEatsMush @goyalayus Can you recommend some top accounts for models as well?

[_arnav_ @arnava_]

@VaibhavSisinty Wondering what ML/DL model has been used, if you guys know something tell me!

[_arnav_ @arnava_]

Implemented RoPE, Causal Attention from Scratch
einsum helping in not messing up dimensions😅