Saurabh Shah's Avatar

Saurabh Shah

@saurabhshah2.bsky.social

training olmos at Ai2, prev at Apple, Penn …. 🎀 dabbler of things🎸 πŸˆβ€β¬›enjoyer of cats 🐈 and mountainsπŸ”οΈhe/him

178 Followers  |  601 Following  |  22 Posts  |  Joined: 26.07.2023  |  1.7392

Latest posts by saurabhshah2.bsky.social on Bluesky

Congrats on publishing this!
I heavily relied on your guide to MLE interviews a couple of years ago - it’s an awesome resource.
I hope you continue with the technical writing and guides 🀠

09.01.2025 18:36 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

+100.
I have way too many mugs. I have ~0 thermos

11.12.2024 01:09 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Is it better?

01.12.2024 23:50 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

oh i didn't realize copilot was literally half the price of the other options. Will need a good reason to get anything else.

01.12.2024 22:05 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

I pretty much only use LM’s for coding, but I’m open to changing that. Which subscription should I buy: cursor, copilot, chatGPT plus or Claude Pro?

Unrelated: how are there no polls on bsky

01.12.2024 21:49 β€” πŸ‘ 3    πŸ” 0    πŸ’¬ 5    πŸ“Œ 0

Attia’s the goat but unfortunately not nearly as famous as Huberman, at least in my circles

01.12.2024 18:03 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Yayyayay

27.11.2024 05:24 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

I love him + I think I have the same fountain

26.11.2024 20:34 β€” πŸ‘ 2    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0
Post image Post image Post image

Philly half marathon

25.11.2024 11:45 β€” πŸ‘ 5    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Huuuge topic, and something I’ve been thinking a decent amount about lately.

Unrelated: loving the new prof pic

21.11.2024 04:49 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Being vulnerable on the TL πŸš€
You got this friend, believe.

20.11.2024 14:34 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Holy moly. Game changer

20.11.2024 14:29 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Good point!

I’m mostly trying to see if there’s any way to increase sample efficiency in fine tuning - in our case this data is coming from annotators so I really want to make the most of it.

The other option that could be cool is making the annotators themselves more efficient

20.11.2024 13:57 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Is there any evidence that data order and repetitions matter (or don't) for finetuning?

19.11.2024 17:30 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

I'm interested in exploring curriculum learning (i.e. being thoughtful about the order and repetitions of data) for text and audio language models.

I'm particularly interested in finetuning, interactions with full tuning vs lora, sample efficiency, etc.

Where do I start?

19.11.2024 17:27 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Wow I really needed to read this, I was just thinking β€œeverything’s so cool how could I pick something to do”

19.11.2024 04:27 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Ok only one person liked this nvrmind im going back to twitter

18.11.2024 22:17 β€” πŸ‘ 0    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Not sure what it is, but vibes here seem excellent. I might stay a while 🀠

18.11.2024 21:35 β€” πŸ‘ 7    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Sure but if Pete asks me why I couldn’t get relora to work I’ll be directing him to you

05.10.2023 18:08 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Is it finally time to abandon Twitter? I was just starting to gain traction there thanks to your qt…

05.10.2023 17:58 β€” πŸ‘ 1    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

Some Saurabh lore: when I made my Twitter account, I managed to get the handle @ saurabh_shah

But I had a typo in my email, so I couldn’t delete my or confirm account

So I did the sensible thing and create a new account under @ saurabh_shah2

I wanted people to find me here, so the 2 remains.

26.07.2023 01:04 β€” πŸ‘ 3    πŸ” 0    πŸ’¬ 0    πŸ“Œ 0

Hello Blue Sky, is anyone out there

26.07.2023 00:27 β€” πŸ‘ 5    πŸ” 0    πŸ’¬ 1    πŸ“Œ 0

@saurabhshah2 is following 19 prominent accounts