HN user

tuhins

95 karma
Posts24
Comments19
View on HN
twitter.com 2mo ago

Waterloo Intern = Head of Infra

tuhins
2pts0
mirage.app 1y ago

Seeing Voices – Generating A-Roll Video from Audio

tuhins
1pts0
www.baseten.co 2y ago

Introduction to quantizing machine learning models

tuhins
1pts0
www.baseten.co 2y ago

SDXL inference in under 2 seconds

tuhins
3pts1
github.com 3y ago

Build a Lensa-like avatar-generation app in a weekend

tuhins
2pts0
www.baseten.co 3y ago

Accelerating model deployment: 100X faster dev loops with draft models

tuhins
1pts0
squidarth.com 3y ago

Teaching Stable Diffusion about Myself

tuhins
1pts0
app.baseten.co 4y ago

DALL-E Mini – Generate images from a text prompt

tuhins
52pts22
math.fau.edu 4y ago

Terence Tao (1984) [pdf]

tuhins
2pts0
news.ycombinator.com 4y ago

Ask HN: What is the fastest way to serve code behind an authenticated endpoint?

tuhins
3pts1
app.baseten.co 4y ago

Demo – Text generation with EleutherAI's GPT-J-6B model

tuhins
1pts0
app.baseten.co 4y ago

GFP-GAN – Photo Restoration App

tuhins
1pts0
blog.baseten.co 4y ago

Transcribing large audio files with wav2vec

tuhins
1pts0
www.baseten.co 4y ago

Hotdog or Not Hotdog (Cutedog)

tuhins
1pts0
blog.baseten.co 4y ago

Working at an early-stage company as an early-stage engineer

tuhins
2pts0
observablehq.com 5y ago

Generative Art using Wind Turbine data

tuhins
2pts0
www.baseten.co 5y ago

Introducing BaseTen — build machine-learning powered applications

tuhins
1pts0
maolson.medium.com 5y ago

Quantum States of the Art

tuhins
1pts0
maolson.medium.com 5y ago

What are quantum computers good for?

tuhins
1pts0
multithreaded.stitchfix.com 6y ago

Stitch Fix: Introducing Our Cultivating Algorithms Visualization

tuhins
2pts1
arxiv.org 8y ago

Classical Homomorphic Encryption for Quantum Circuits

tuhins
1pts0
news.ycombinator.com 8y ago

Ask HN: How did you organize your technical job search?

tuhins
12pts6
karthkk.wordpress.com 10y ago

Deep learning solution for netflix prize

tuhins
5pts0
blogs.princeton.edu 10y ago

AI and the current understanding of deep nets

tuhins
4pts1

I was one of them! Joined on August 5, 2012, got 0.5% of equity (I think?), it all went to more-or-less zero monetarily. I don't think most of us really hold any grudges against Sahil here. It was a very fun place to work, and I met some of my closest friends and made some of my strongest professional connections there. It was net-very-positive to my life and career, and I think we were all adults when we were opting-in to the experience.

As for Sahil/Gumroad making money and growing. Meh. He's worked on it for 13 years and showing dedication beyond what I would have for most things. It's fine.

We haven’t hit the invoicing issues that OP describes, but any form of MRR reporting is a surprising challenge. Discount attribution is really hard.

No-one cares if you fail and in the absence of unethical behavior, your identity and your company's identity are distinct. As a founder you think the world has its gaze on you and inflict enormous amounts of pressure on yourself. If things aren't going well, it's completely fine. If it's time to kill it, even better. Fail fast, fail often, and take care of yourself so you can run it again.

I’m Tuhin, one of the co-founders of Baseten. Baseten is a realization of the product that we wish we had when we were trying to get value from trained models in previous jobs. I’m excited to see this on HN and will be around to answer any questions you have.

This is an interesting question. I think the software industry (speaking from someone who works at a startup in SF) today puts a high emphasis on "culture fit" which is best determined (imho) by committee. I think you'll find that other industries (banking, consulting, law) also put a lot emphasis on hiring by committee; one hypothesis I have is that these industries, like the tech industry, are high churn industries (average tenure ~2 years), so if your champion leaves, you'd best hope everyone else wants you to be there.

This is pretty awesome. I think what's as important as actually providing the data + results is details of the steps taken to produce the analysis. Clinical trials data analysis is often performed by underqualified research assistants and then sent to a biostatistician for a rubber stamp. Having a mechanism to review analysis would not only save time in getting potential treatments to patients, but also reduce the effort required to analyze new datasets.

If you are intelligent, have opinions and can justify your opinions, most people will take you seriously. Whether its a mistake or not depends on you. Are you completely set on working in tech? College for most is a time for exploration and you will miss out on this. Also, not having a degree precludes you from many careers outside of tech (investment banking, consulting, professional careers in law, medicine, etc). That being said, if required, you can always go back to school later.