Read on Twitter
I almost run into someone with a few years of technical experience who has questions about ML Careers or Startups
So here's some off the cuff notes friends --
So here's some off the cuff notes friends --
Big Co vs Startups: Odds of startups going bust is less than you might think. In my 4-5 years I've never seen a Series C startup go down in India
Raising VC money, building distribution and an engineering team are so damn expensive in India -- they're a competence test of sorts
Raising VC money, building distribution and an engineering team are so damn expensive in India -- they're a competence test of sorts
Big Co vs Startups: Most startups pay within 10-30% of your BigCo company, specially engineering and so typically the cash hit which you would have to take it less than you might think.
Big Co vs Startups: Paradoxically, if you're coming to startups to learn ML -- this is the stupidest thing I've ever heard. Learning ML needs mentors who know this stuff AND have the time to teach it
Even if a handful of startups have the talent, they rarely have the time
Even if a handful of startups have the talent, they rarely have the time
Big Co vs Startups: If you are thinking that startup equity/ESOPs are gonna make you rich, you're better off playing the lottery.
Finding someone who has bought a home from vested stocks is rarer than finding a unicorn
in India
Finding someone who has bought a home from vested stocks is rarer than finding a unicorn
in India
Big Co vs Startups: As an engineer, if you're not doing interesting work at a consumer internet company in India -- you're doing something wrong.
Whatever you're doing wrong there, you'll do it in your startup engg career too. There is fuck-ton of more interesting work in BigCo
Whatever you're doing wrong there, you'll do it in your startup engg career too. There is fuck-ton of more interesting work in BigCo
On ML: I can think of maximum 2-3 of businesses who would really miss their ML teams if they all died tomorrow in a freak bus accident tomorrow.
The net business value added for most seed to Series B startups ML team is probably negative
The net business value added for most seed to Series B startups ML team is probably negative
ML is MOSTLY Engineering:
Most startups don't need anything more complicated that a finteuned BERT or ResNet. That is when they've a proprietary dataset -- most of us don't.
You'll spend 95%+ of your time doing deployment, data wrangling, and not fancy RoBERTa experiments.
Most startups don't need anything more complicated that a finteuned BERT or ResNet. That is when they've a proprietary dataset -- most of us don't.
You'll spend 95%+ of your time doing deployment, data wrangling, and not fancy RoBERTa experiments.
Startups don't love you:
If you don't have a PhD and work as a Data Scientist -- I've a surprise for you --- as they scale, they will rather hire someone with a PhD to "supervise" you than to get you 100 hours of 1-1 coaching.
This is for ML engineering roles too.
If you don't have a PhD and work as a Data Scientist -- I've a surprise for you --- as they scale, they will rather hire someone with a PhD to "supervise" you than to get you 100 hours of 1-1 coaching.
This is for ML engineering roles too.
Most of what you know in terms of specific ML tech is going to get outdated in less than 4 years.
This usually means that someone who graduated this year is factually more effective than you at ML modeling than you because you've not read a single damn paper in last 6 months
This usually means that someone who graduated this year is factually more effective than you at ML modeling than you because you've not read a single damn paper in last 6 months
90% of all ML projects fail. The reasons vary but it's almost never because your ML "model" sucked.
It's almost always because the business, product and you have no idea what you're doing and pretend otherwise. Have some fucking humility and ask for behavioral test cases.
It's almost always because the business, product and you have no idea what you're doing and pretend otherwise. Have some fucking humility and ask for behavioral test cases.
Data Scientists *need to know* how their work ties into business or buyer value. If you don't and work at a startup, you're playing cricket with one hand and one eye.
Please be kind to yourself and learn what the heck actually makes money in your business. Go join @StoaHQ maybe
Please be kind to yourself and learn what the heck actually makes money in your business. Go join @StoaHQ maybe
The other biggest reason you are going to suck as a data scientists because you don't treat the DATA like it pays your salary.
Please treat it with the sacredness of a scientist and archaeologist and not like an engineer's playful stupidity of fancy new DBs
Please treat it with the sacredness of a scientist and archaeologist and not like an engineer's playful stupidity of fancy new DBs
The biggest bottleneck in your ability to add value is going to be your inability to read the fucking docs and/or inability to do any DevOps at all (guilty as charged of second)
The full stack data scientist needs to know how to take the same damn standard BERT/ResNet to prod!
The full stack data scientist needs to know how to take the same damn standard BERT/ResNet to prod!
The strength of technical innovation isn't gonna come from training your models by reading a damn Medium article.
Go learn a bit about confidence intervals, data efficiency, (test time) augmentation, weak supervision, avoiding over-fitting and how to explore weird datasets.
Go learn a bit about confidence intervals, data efficiency, (test time) augmentation, weak supervision, avoiding over-fitting and how to explore weird datasets.
Legend has it Edison had to experiment with 1000 substances to discover the carbon filament for his bulb
. But he already knew that the material will need a high combustion temperature, melting point and high electrical resistance.
All innovators get paid for EARNED SECRETS.
. But he already knew that the material will need a high combustion temperature, melting point and high electrical resistance.All innovators get paid for EARNED SECRETS.
Don't be a jackass and go ahead and write a blog about your EARNED SECRET. Take it to the last mile that you can and then write about it.
Contra: Anything that you learn from a blog, can be taught to anyone. What makes you so worthwhile hiring?
Contra: Anything that you learn from a blog, can be taught to anyone. What makes you so worthwhile hiring?
The differences between different managers and teams at a company can easily be larger than the differences between companies.
Whatever someone is telling you, that's usually saying that they had a great manager in that career track. Including Founders and VCs
Whatever someone is telling you, that's usually saying that they had a great manager in that career track. Including Founders and VCs
Okay. I think I'm done for tonight. Will pause and maybe make some more tea 
Here's the start of this thread in case you wanna RT/share or something.
Don't bookmark, cos this will get deleted in 7 days https://twitter.com/NirantK/status/1324416922960625665
Here's the start of this thread in case you wanna RT/share or something.
Don't bookmark, cos this will get deleted in 7 days https://twitter.com/NirantK/status/1324416922960625665
