cweill

100% agreed. Start with the $10 plan. I forgot how much it cost for the middle tier. The one benefit of the higher tier is you get access to better GPUs, and can run multiple colabs in parallel effectively getting multiple accelerators at once if you're doing distributed hyperparameter tuning.

Thank you! Feel free to reach out to me on Twitter or Discord (linked on the homepage) if you want to chat.

I've seen this chart too. But research != Production. I bet that TensorFlow is still more commonly used in serving than Pytorch because of great tooling like TensorFlow Serving. I could be wrong though, as I'm not up to date with the latest in the Pytorch ecosystem.

Not too late. You've probably derisked the product side a but since you already have years of experience building technical products. You just need to learn the basics of running a business now, and get comfortable with sales and marketing.

I'm 32 and just started a solopreneur bootstrapped business. I have much more confidence now than when I was in my 20s.

The biggest thing is that I'm more patient now. In your 20s, it's easy to feel like you need everything right now. But as you get older, you learn patience. I'm still impatient at times, but I also feel more comfortable saying I'll spend the next 4 years trying things, than trying to race to success and giving up at the first sign of trouble.

From what I've seen, the people who start successful businesses in their 20s are from generational wealth and can afford the career risk, or their parents are entrepreneurs. Then you also have the people who have been coding since they were 8 years old, so by the time they are 20 already have 10 years of experience. Plus, code has the wild scalability of infinite permissionless leverage (as does social media), so it's possible to make millions from your bedroom as a high school or college student.

In short, give it a shot for a year. If you hate it, you can always go back to a job to learn from your mistakes on someone else's dollar!

I worked in Google Research for over 5 years doing Machine Learning, and recently quit to build my own ML start-up. These days, I solve a mix of NLP, computer vision, and tabular problems, all with state of the art neural network techniques. I've tried many setups.

My advice is go with Colab Pro ($50/mo) and TensorFlow/Keras. You can go with Pytorch too if you prefer.

I made the mistake of buying a 2080Ti for my desktop thinking it would be better, but no. Consumer grade hardware is nowhere near as good/fast as the server grade hardware you get in Colab. Plus you have the option to use TPUs in Colab if you want to scale up quickly.

You really don't need to get fancy with this setup. The best part of using Colab is you can work on your laptop from anywhere, and never worry about your ML model hogging all your RAM (and swap) or compute and slowing your local machine down. Trust me, this sucks when it happens, and you have to restart!

As for your data, you can host it in a GCS bucket. For small data (<1TB) even better is Google drive (I know, crazy). Colab can mount your Google drive and loads from it extremely quickly. It's like having a remote filesystem, except with a handy UI and collaboration options, and an easy way to inspect and edit your data.

Ever done machine learning? Chasing that ever improving accuracy is a high like no other. It's random variable reward. Ask anyone who does ML for a living.

Posting one thing on Twitter every day that I think will be valuable to someone. It's part of showing your work, which is important when building a bootstrapped solopreneur business. I've met clients and valuable connections that way.

A few things (as a user of Google's OCR API):

* How does this compare to other OCR tools on standard benchmarks? * I don't see a terms of service, you might want that. * The privacy policy also seems to be lacking. IANAL but you should hire one to make sure that stuff is squared away.

Just like CollegeACB and EzraHub from back in my day.

I don't see the different between marketing and customer development à la Lean Start-up. As a budding solo-developer myself (2 months in), I'm making sure to get my customers involved from day 0: speaking to as many as possible, learning about their pain points, and only building one feature at a time before reaching out again. Some of your most loyal customers will be along for the journey.

Twitter has been huge for me to find my customers. I'll just DM people who talk about relevant topics and tweet once a day about what I'm working on. I just got my first paying customer (a big one too) through someone retweeting one of my tweets. At the end of the day your customers are people too.

I agree that marketing is scary, and a good book about this is "Show Your Work".

A big part of the problem is that a lot of success you see is due to non-linear outcomes with high upside. However, we humans are predisposed to thinking linearly, I.e 2x the effort leads to 2x the outcome. This cognitive dissonance makes us unhappy.

For example anything produced with code or most digital media have these effects. Your Tycoon game example is this: it seems like such an incredible thing for one man to not only accomplish, but also for you to have discovered and enjoyed it.

If you apply linear thinking when comparing yourself to someone with non-linear output, you will feel like you could never reproduce that in 10 or 100 lifetimes. And you would be right! But bear in mind the only reason they could have that success is because they chose to work on something with a non-linear outcome. And the only way to be like them is for you to chose to expose yourself to similar upside, and the risk that comes with it.

For more on this topic, read Taleb's Antifragile (and other books in his series). Reading his books led to a paradigm shift in my mind that made this way of thinking no longer an issue for me.

Most of the stories I've read online have been positive about their land. I recently left my job at Google, and yours is a sobering tale that one should not take their comfortable job for granted. I'm giving myself 1 year to see if I can manage on my own, but if that doesn't work out, I'll probably return to a FAANG company.

I do believe that those kinds of jobs are the best risk-adjusted ROI if you're trying to make money. But sometimes you just need to get the startup out of your system.

Same here. Too bad it only lasts 6 months.

Can you please write a blog post or book about your experience? I'm trying to get into solo SAAS, and this kind of post is pure gold for me!

Sightly OT: the best Math for Computer Science book/course I ever read is the one from MIT https://ocw.mit.edu/courses/electrical-engineering-and-compu...

Prof. Tom Leighton is an incredible teacher who teaches the intuition before the theory.

Alfredo is an excellent teacher who really cares about his students learning. I've watched a number of his lecture recordings on YouTube, and would highly recommend this course.

Agreed. ROC AUC is fairly robust to over or undersamping a class in determining if your classifier is predictive.

Please make a follow up post. I’d love to learn from your experience.

Speaking as someone who just quit their FAANG job to go out on my own, your comment is inspirational!

Thank you for your transparency.

Thank you for this book. It's the best book I read all year about figuring out if the thing you're building is a complete waste if time or actually solves a real need.

I used its guidance at my big tech job when trying to build a new tool for researchers, and it helped me better understand how to understand what they need, iterate, and better tailor the sales pitch afterwards. Who would have thought that talking to customers is the single most valuable thing you can do!

I had the same issue and same realization. That's why I always use keyboards without the numpad; that extra 4 inches of space for the mouse relieved most of my computer posture pains.

Great execution, and very neat app!

But, what's wrong with using Amazon affiiliate links? If anything, monetizing would be great since it would give you more incentive to maintain this wonderful application? And it doesn't cost us users anything.

The key is that step 1 isn't random: it's guided by past experience, and intuition. At the local level it might be random, but from a higher level, the search is guided.

That said, there are incremental papers that tweak some parameters, and then there are others like this one that take a big, risky step back from the rest of the community (I.e using convolutions for computer vision), and make a discovery on the common benchmark.

Speaking my own experience, I think what keeps researchers addicted is the random variable reward we get from seeing our new algorithm's performance on a common benchmark. Even better if the performance near state of the art, but not necessary.

Congratulations! What kind of corporate structure do you use for your microbusinesses? Delaware LLC? Also are you based in the USA?

Yeah WSL 2 is amazing. You can even connect to your Nvidia gpu from there Linux subsystem if you want to some local ML training on it. Goodbye dual boot!

Perhaps that's true. A startup will put your friendship through a crucible, and at the end, whether success or failure, if they are still there for you, you know they were a true friend. And true friends are the people you want to go into business with. We need to think of startups as a repeat game with multiple iterations. The optimal strategy is to cooperate, not defect. Your first startup does not need to be a success, but a great tool for finding worthy collaborators.

What's well paid for FAANG senior? 300k? 400k? $500k+? Are levels.fyi's numbers out of date?

As much as people love Napoleon Hill and his book, he may have been one of the first and biggest fake gurus of the 20th century, having never actually interviewed Carnegie and other rich people for his beloved book: https://youtu.be/iEDzb9ZplX0

Did you mean "survive"? Because I enjoyed the distopian pun of "servive": living to serve to live to serve.

I don't see golang on that chart. It must be close to C (on the lower end).