"even for highly technical aspects like deep learning, fast.ai has shown that people with 1-year of coding experience can become world-class deep learning practitioners"
Yes, any Joe can train deep network with dozen lines of Keras. Sure, your startup can use off-the self models and tweak it a bit... That doesn't make you "world-class deep learning practitioner". If you are designing the network for new problem domain, there are thousands of decisions to make - everything from hyperparameters to network architecture to distributions in data. Making these decisions without having developed lots of intuition and good foundations is very hard. I often find usual developers without ML background and experience completely lost in these cases. Debugging a model that doesn't work is super hard. There are no IDEs, no breakpoints, no watches and in fact not even error messages. Its purely statistical debugging and probabilistic fixes.
>Startups are (by necessity) filled with generalists; big companies are filled with specialists. People underestimate how effective a generalist can be at things which are done by specialists. People underestimate how deep specialties can run. These are simultaneously true. [0]
For Google etc. it probably makes sense to pay top dollar for lots of ML PhDs (specialists).
A startup looking for opportunities engendered by ML is better off with a smart generalist. E.g. Dawson Whitfield of Logojoy was a designer. [1]
Yes, because at their scale and revenue, it makes sense to start pumping out your own research. Some PhD theses do eventually become competitive advantages (Paxos comes to mind), and nothing is better than being able to work on your “thesis” with a nice paycheck and without the threat of “publish or perish.”
World class is subjective. The term here is practitioner and the simple answer is yes: you will be world class because so few people can do it. World class researcher? No.
Reading HN and following key people gives you the impression that this is an incredibly popular field - it is. In reality, though, it's a small field and lots of companies still don't use it, intend to use it, or have any idea how to use it. Alexnet was only 5 years ago. Faster-RCNN was only 2 years ago. Compared to, say, all programmers, statistically nobody is doing this.
Debugging a model is hard, but it's unlikely your efforts will be significantly more or less productive than a leading researcher aside from obvious things like noticing that the learning rate is too high or low. There's so much hand waving, even in top papers, that mostly we've tried stuff that looked like a good idea, and it turned out to be correct.
The thing is that a lot deep learning tasks don't require a fancy model. You can take VGG/Inception/ResNet and bodge them to fit a lot of real-world scenarios. Then the problems are mostly solved by standard machine learning intuition - making sure you have an appropriate train/test/validation set, a good loss function, etc.
Edit -
Here's an example: Halcon, a popular commercial image processing library has recently included a deep learning package due to repeated customer demand. It's not out yet, but it's most likely just going to be a wrapper for train/test using a basic image classification or object detection network. That's what the market wants. They don't care about GANs, they want something which will tell them if the image represents a valid product during QA 99% of the time (and often QA failures are pretty obvious - e.g. big crack in a bottle, or a deformed label).
You could build that kind of model with very little expertise and you'd beat most of the available image processing packages easily, with practically no work. The point is that an off-the-shelf model trained on company-specific data is likely to be state of the art on that problem (because who else has the dataset?) You don't care about 1000 classes in Imagenet, you want good product/bad product.
If you spend one year on applying deep learning, you can train a net on a 100 different data sets. That's where the intuition comes from. You'll debug a lot. People with zero experience with deep learning have ended in the top 10 for Kaggle computer vision competitions. Applied ML does not require a PhD in Computer Science. It is not like you are debugging Cuda as was the case a few years back. Even then, a PhD in statistics is unlikely to help with that. Really, one year of Python experience and you are good to go.
How rare is it to find a new problem domain these days? Most if not all problems tackled with deep learning have already been done in one form or the other. Deep learning is not yet much of an academic discipline, but accessible to anyone willing to put in the hours of study and practice.
"How rare is it to find a new problem domain these days? Most if not all problems tackled with deep learning have already been done in one form or the other."
Curious if anyone's done this: deep learning on DOM trees, to recognize portions of HTML pages. This has elements in common with both CNNs (where you want surrounding context - parents, ancestors, siblings, and descendants - to influence the recognition of a particular DOM element) and with RNNs (because elements may have an unbounded number of children, and potentially a recurring substructure), but I don't know offhand how you would combine them to solve the problem.
Anyone working on this? A quick literature search turned up nothing.
Nothing is unlearnable - it's never question of "if it can be done" but question of "how much effort is needed". Read few blogs, books and take Coursera classes etc are all good but the real thing is working on multiple real and new problems. Anyone who has read books/blogs etc, give them a new problem, for example, predicting pedestrian intent for crossing a road from set of video frames. See how far they get.
So are you saying there are no best practices or patterns that are discovered over time that allows newcomers to progress in the field faster than the early pioneers?
Sorry to be that guy but this kind of "elitism" seriously bothers me. I've seen this thing in almost all of the fields related to CS like "you have to spend 16 hours a day for 10 years to understand assembly" when I know a guy who became a malware expert in 4 years ( winning world class tournaments with his team ).
This thing is very discouraging, and it is false. You stated a problem which is hard to solve, but 99% of problems are not like this. Like:
- tell what kind of mineral deficiency a plant has?
- what kind of bug do you see?
- what kind of plant do you see?
- is this thing broken ( MRI or X ray )?
- what kind of musical notes are in a waveform?
See how far they get. A talented guy can solve these within one year.
I read that in the sense that you don't need very many ("a bunch") of PhDs in a small company. That does not imply that one or a few people of that background might not be useful.
If AI is an integral part of my startup I would prefer to have someone with adequate experience in the founding team or very early hire.
90% of the work might not be super deep fundamental design decisions or debugging.
"1-year of coding experience" for a startup means someone who got a PhD and interned for 4 years at various companies doing real work, then went to work for Google for 1 year and decided to leave for a smaller company.
Yes, that guy can become a world class practitioner.
He was talked down to junior and paid as little as possible because that's what companies do.
I don't want to diminish the success that this company is having being run in their own way, but give me a break with these blog posts.
VC is a "trend to avoid"? Avoid "Hypergrowth"? I guess companies like Google, Facebook, Twitter, Amazon, Stripe, Airbnb, Dropbox, Pinterest and a hundred others all really messed up.
I mean I could go point by point and give examples that are the opposite of these trends, but that isn't any more meaningful than these suggestions. You can choose to run a company however you want, and you should. If you believe these things are bad, don't do them. Attract employees that agree with you. The market will decide if you are right or not.
But enough with the blog posts about it, especially if they are dismissing critical tools that many companies use very successfully. Every company is different, every product is different, every team is different. Use some common sense on what tools make sense for you, and keep an open mind.
Counterpoint: Blog posts like this one are useful, because the VC-funded companies you named are so successful at dominating our thoughts, conversations, and news feeds (HN included) that many founders never see any examples of people following an alternative path.
It can be inspiring to a lot of people to simply hear about the possibility of building a small, revenue-generating business or side project.
I've been running Indie Hackers for the past year and a half, interviewing hundreds of developers about their small businesses[0], and it feels like every day I talk to someone who's never even considered bootstrapping to be an option. Spreading awareness is valuable, especially considering that venture capital is only a viable option for a small minority of companies, and the ways in which it can cause an otherwise healthy business to fail are rarely discussed by VCs themselves.
>because the VC-funded companies you named are so successful at dominating our thoughts, conversations, and news feeds (HN included)
Right - HN's corporate sponsor is YC Combinator, so it's kind of inevitable that the VC narrative dominates.
The good thing about Indie Hackers being acquired by Stripe in April [0] (congratulations!) is that, even though you now have a corporate sponsor too, in the short term at least nothing really changes.
You wrote that
>Stripe wants to grow the GDP of the internet
so having 1,000,000 Indie Hackers starting companies is completely aligned with Stripe's corporate interests and that at least provides a counterpoint to HN and YC Combinator.
Further out, if Stripe Atlas grows big enough then we could end up with an Amazon Marketplace situation where small companies are forced to take whatever terms "Darth Vader" (#patio11) offers.
But we're a long way from that point yet.
A 3rd way is not to start a business at all. As Vincent Woo said in his recent podcast with you [1]:
>Right, so readers at home, if you can start a business, do that, but also maybe don't. It's not easy, it's a lot of work and there are a lot of things that valuable in life that have nothing to do with money, that's how I'd put that.
> Right, so readers at home, if you can start a business, do that, but also maybe don't. It's not easy, it's a lot of work and there are a lot of things that valuable in life that have nothing to do with money, that's how I'd put that.
I hate this narrative.
Yes, there are things in life worth more than money, but it just so happens that having a shitload of money makes living the life you want WAY easier.
What Vincent is saying is “Doing startups is hard, so just settle and be happy with what you have.”
What if you’re not? It just so happens that making very large amounts of money as an engineer is hard too!
Oh indeed, but seeing as we work on HN every day, we're more steeped in the data than anyone else is, and to us this is a pretty obvious observation. I'm saying 'we' because I'm sure all the moderators agree.
The HN community stopped being based in SV many years ago, if it ever was to begin with. It's heavily international, widely distributed even in the US, and significantly more sympathetic to boostrapper/indiehacker narratives than to VC ones. Which is pretty easy to understand if you think about who the users are and how many of them there are.
In terms of comment quality, the cynical end of anti-VC sentiment is a problem because it's so knee-jerk and predictable. Solid critique is welcome, but alas, there isn't much. As a moderator I don't care what people like or hate but I do care about getting the best insights from all sides, and this is one area where we don't, really. Or rather, we do, but it's drowned out by Dunning-Kruger and reversion to the mean and all the other phenomena that make internet forums suck.
> it feels like every day I talk to someone who's never even considered bootstrapping to be an option
It's mind-boggingly to me that this can be an option that people have never considered but I totally believe you. I've seen people have the "epiphany" that companies can spend revenue in much the same way they spend VC money. I guess their model is that only after that money is distilled into an up-and-to-the-right graph and put in front of VCs and they give you money do you actually get something you can spend.
I disagree that this blog post is giving an interesting counterpoint. As I said, I'm not diminishing their success, but this blog post isn't about their story, it's about knocking other styles of companies they disagree with. For example, the VC thing:
> Therefore, VCs often push companies to grow too quickly, before they’ve nailed down product-market fit and monetization
[citation needed]
> Staying small keeps you focused on a small number of high-impact features.
[citation needed]
> This is not just a few bad actors: the behavior is wide-spread, including by many well-known and ultra-wealthy investors
[citation needed]
-------
I'm not going to get on some high horse here defending VCs, but these aren't original lines of argument and when unsupported I'm not sure what value they are providing. For example, YC is a VC. Do you think these 3 things apply to them?
Again, good for this company, I'm really happy they are succeeding and I'm glad they are doing it without VCs or other trends they don't like. This blog post is very shallow marketing though, I don't see how this is adding to the accumulated knowledge of creating a startup.
I don't think a citation is needed for the proposition that VCs often push companies to grow too quickly, before they've nailed down product-market fit and monetization.
You could argue that VCs want to see product-market fit before they invest, but this is certainly not true of monetization.
> VC is a "trend to avoid"? Avoid "Hypergrowth"? I guess companies like Google, Facebook, Twitter, Amazon, Stripe, Airbnb, Dropbox, Pinterest and a hundred others all really messed up
I'm not going to say that VC isn't ever or usually the right move, but this comment is the height of survivorship bias. Not every business idea should be a VC-funded startup and trying to shoehorn every team and every business model into that path is a mistake. Whether to take VC capital is situational and depends on the team, market, product idea and all other factors of that specific startup. The businesses you listed were all the proverbial winning lottery tickets. There's plenty of losing lottery tickets too, and many of those had great ideas that would have been better served with a bootstrapped model that looked for slower, more sustainable growth and increased founder control.
A great startup is like a great food recipe. You have to mix the right ingredients in the right amounts to make it taste good. Almost every ingredient works in some dish, but many ingredients don't work in combination. Dishes can be appetizers, mains or desserts, and you'll want to add different ingredients depending on which one you're making. VC is like sugar. It can make your startup fat in a hurry, but if you're growing very quickly, all those calories can keep you from starving. It tastes great in almost anything, but it's put into way to much of what we eat today and causes a lot of premature deaths.
Another useful metaphor is that venture capital is rocket fuel. It's not useful if you're not building a rocket. And VCs only really want to give rocket fuel to rocket builders, anyway. VCs will be angry if you end up building a car or a plane instead of a rocket, no matter how awesome the car or plane that you built is, because they promised their investors (LPs) that they would only invest in rockets. But if you've clearly already built a rocket (perhaps after years of bootstrapping, and it's now obvious you can reach orbit -- if only you had enough fuel -- then venture capital, err rocket fuel, might make sense.)
It was interesting reading this, because those 5 points basically describe Google to a T. It's an aborted Ph.D thesis that took VC, hired lots of Ph.Ds with a culture "like a family", and then embarked on a hypergrowth trajectory.
The irony is that when Google was young, there was a different set of "destructive trends". They still included taking VC, but it was also common knowledge that:
1. You had to sell a physical product for money.
2. You had to build out all your infrastructure before you could bring a product to market.
3. Sales & marketing were more important than engineering.
4. (When they were hiring on their hypergrowth trajectory, 2001-2004:) The web is dead, and programming is a terrible career because it's all about to be outsourced to India.
Perhaps the meta-lesson is to do what other people are not doing. The point of markets is that they reward unexpected successes, where an entrepreneur serves a population with little competition, and that necessarily means thinking for yourself and ignoring the common wisdom. Fast.ai does an admirable job at that - their deep learning course is excellent - but taking their advice at face value is just as dumb as cargo-culting Google.
> Perhaps the meta-lesson is to do what other people are not doing.
A better lesson might be: stumble upon a phenomenal river of gold and dive in head-first.
Unfortunately, rivers of gold are rare. Which is why the rest of us are stuck with mundane stuff like marketing units of product or service to sell at a profit.
Even a broken clock is right twice a day! These companies had awesome market fit, and also lots of luck. Thousands we don't remember have tried this and failed.
Hypergrowth/VCs have been oversold as 'The one way'; I'm glad to see blog posts showing that there is a whole world out there.
The problem with VC money is that you are likely taking on powerful investors / co owners that are unlikely to have the same goals that you have (or should have).
VC money comes from those that have already become successful. So they are pushing for a strategy that will cause the majority of companies to fail but with the small chance that one will become the next Google, fb, etc.
Most founders can't afford that methodology. They should be looking to build a good company with good cashflow that can earn them a healthy nest egg and make them wealthy. When was the last time you heard of a VC that wanted good dividends from their portfolio companies?
So yes, if your goal as a founder is to build the next unicorn and you are willing to risk everything on that small chance to make it super big, then you should go the VC route. Problem is, you have one company and one life, so you better hope your company is the one in 10 that have a profitable exit.
Or maybe save yourself some time and just go buy lottery tickets.
I think VC-funded hypergrowth raise your company's chance of failure. However, it also raises the chance of a big exit. Hence, high risk, high return.
VCs are in the business of investing in high risk, high return companies.
So it's natural they want their portfolio companies to optimize for high return than survival.
If any one of the VC's 50 portfolio company makes a big exit, the VC profits.
For founders and employees, you can't work for 50 different high risk, high return companies at the same time, so that might not be what they want.
I feel like the expectation after getting VC money is "do it fast" to the detriment of "do it right". We may have had the wrong VCs or CEOs where I have been, or maybe I brought my own bias, but that definitely felt like the push.
Part of the m.o. for many VCs is to push for really high burn rates. Which has two reasons, one is to accelerate growth as you mentioned. Another is to cause you to spend the funds you raised and require more investment sooner. This giving the VC the option to own a larger and larger share of your company by the time that liquidity event hits.
To be honest, Twitter seems like exactly the kind of company that this advice was made for. It had a neat idea that could have been made them a steady profit as a small to medium sized company, but took on billions of dollars in funding, hired thousands of employees and hired a bunch of scientists and high end engineers for a project that arguably didn't need most of that stuff to begin with.
Hence what could have been a decent company making a cool few million a year is instead one making huge losses on the assumption it's the next Facebook.
> I guess companies like Google, Facebook, Twitter, Amazon, Stripe, Airbnb, Dropbox, Pinterest and a hundred others all really messed up.
Google and Amazon are hardly the classic 100x-hypergrowth model.
Facebook and Twitter DID really mess up, have you seen how much utterly vile stuff you can report to them and it will still be around months later? Have you been at the receiving end of a shitstorm or Nazi trolls? People order SWATs over Twitter, leading to deaths. The president of the US leaves the rest of the world in panic when someone will finally declare war with him after yet another tweet. The reports from former "moderators" all have one thing in common, they're exploited worse than cannon fodder in wars (the soldiers at least were dead, the moderators will have to live with the videos of beheadings and CP the rest of their lives). Airbnb screwed up entire city rental markets and squeezed out thousands of poor people, hardly call that a success. Pinterest? Wtf how on earth is this worth 12 billion dollars? Or Snapchat with 25B $? Where is the actual worth represented in positive effects for society by this?!
In the end all that creates this "wealth" is data. Data is volatile, it's intangible - and once it is too much, the noise ratio gets too high to be useful, and then most of the companies you mentioned will have a massive problem with finances because they're only worth as much as the data and eyeballs of their customers is worth.
> Attract employees that agree with you
and end up like Uber with a reputation of "every one is welcome here, as long he's a white male with tolerance for extremely high alcohol consumption"?
> The market will decide if you are right or not.
The market is so oversaturated with cash that an app that could literally do nothing more than distribute "Yo" could get 1.5M$ cash at 10M$ valuation. The corrective forces of the market have vanished long since, otherwise the business model of many VCs would have broken years ago. To make stuff worse, the employees are often enough paid with options that may or may not become entirely worthless (and some options even are until an IPO or after years of working for the company!). That's gambling with the future in a really toxic way.
And while I'm at gambling: that more and more cash, both from the VC, the institutional financial and the private sector is flowing into ever more "coins" is also something that is dangerous and set up to explode. Yet another consequence of the current cash saturation.
Don't get me wrong: VCs as an institution, startups and coins serve an useful purpose. But in current conditions (both political and financial) they and the business model they run have grown extremely toxic on society.
> The president of the US leaves the rest of the world in panic when someone will finally declare war with him after yet another tweet.
You know if people are actually in panic over him tweeting, maybe the problem is in those people.
It's like those folks posting pictures of their kids crying when Trump won, saying "my 6 year old is now scared he'll be murdered". Uh, you did that to yourself.
I have a vested interest as a VC, but I disagree that VC funding is a negative signal. That's a very blanket statement to make, and there are lots of toxic VC-funded companies but also lots of amazing ones. (Similarly, there are tons of toxic non-VC companies and tons of amazing ones.)
The way I'd frame it is:
1) If you want to build a company that is venture scale, taking VC funding is a great option to consider. If you don't want to build such a company, VC funding is a very poor option.
2) There are good and bad VCs. The bad ones suck. The good ones will help you and support you even if the outcome is 2x or 0x. Fred Wilson at Union Square Ventures articulates this attitude well in several posts:
a) "If you look at the distribution of outcomes in a venture fund, you will see that it is a classic power law curve, with the best investment in each fund towering over the rest, followed by a few other strong investments, followed by a few other decent ones, and then a long tail of investments that don’t move the needle for the VC fund.
But that long tail is comprised of entrepreneurs and their teams. People who have given years of their lives to a dream that was ultimately not realized.
And as I have written many times over the years on this blog, I spent the majority of my time on that long tail. This is irrational behavior if you think about fund economics, but I believe it is rational behavior if you think about firm reputation." (http://avc.com/2015/11/power-law-and-the-long-tail/)
b) "There are two interesting things here that I always think about. The first is that even the very best investors in the VC business only get a hit about 1/3 of the time. That means that they have their share of "slog it outs" and "hit the walls" too. I am certainly in that camp. The second is that we end up spending an incredible amount of time and energy (hopefully not money) on the 2/3 of our investments that don't work out. When everything goes well, you really don't need that much from a VC. Of course, I have added value in all of my winners. But its the ones that don't work that I have left my blood, sweat, and tears on. And that's the paradox of being a VC that cares. Which is the only kind of VC you want to work with."
(http://avc.com/2013/03/when-things-dont-work-out/)
I'd frame 1) slightly differently, as there are many cases of bootstrapped unicorns.
Time is also a currency, so you make a trade-off between time (extra years of your life spent acquiring funds to scale, and delayed product launch) and control.
Time is also an issue when you have competition, in the sense that your competitors can get there first by raising money.
I agree with you. I tried to be deliberate about this by saying VC is a great option to consider for venture scale company -- but it's definitely not the only option.
The common mental model of a fledgling software startup is too large by at least one order of magnitude.
For a decent, traction-but-no-rocketship product oriented startup, you need one solid back-end+ops guy, one solid front-end+UX guy, and one programmer for each mobile platform you wish to support. Add a CEO+sales+finances person, to keep the business side of the business compliant.
Yes, you won't get a cool continuous integration autoscaling whatnot doodad. YAGNI also applies to infrastructure, people and organizational hierarchies. Don't build them just because Google and Facebook have them.
Look for a profitable company that's just one size bigger than you are today, and aim for that. They've already proven that it's possible to operate at that size with the whatever they've got. Repeat as you grow.
Yep, all the points are the truth. I've got some of those mistakes in the past especially VC funding, and it's really painful. Staying small is not a big lose or something, actually you can earn 10x more money by staying small than burning billions and keep working crazy hours. For example you can look on what's up for the price of 19 instagrams they had around 30 employees. They were small and output was way bigger than in general VC - funded company.
Great points. Strongly agree with #3 in particular ("like a family") because I've made this exact mistake with companies I've led in the past.
To elaborate on what the piece touches on but doesn't specifically say:
> You will need to make hard decisions for the sake of
> the business. You can’t actually offer people anything
> remotely close to lifelong loyalty or security, and it’s
> dishonest to implicitly do so.
To be clear(er): You will have to fire people, and firing someone who thinks of themselves as a family member or who you think of as similar makes the whole thing much more painful. Further, it can make you, as a leader, hesitate when it's an action you really need to take ("but this person is like my brother - we'll make it work!").
At my most recent company, we took the opposite approach -- we all liked each other a lot, we worked well together, and we ate lunch together as a team, but at 6pm everyone went home to their own lives and families. The lines were clear, the understandings were there, and I think it was a much better way to run things.
Hmm, this article took a quick turn into a story about how it takes just one year of coding experience + something from fast.ai to become "world-class deep learning practitioners"; make of that what you will.
I don't see anything that mentions just a year of coding experience, but as a participant in their free MOOC I can say that it's highly worthwhile. It's unique and complimentary to the more academic material usually found on deep learning.
From the article... "And even for highly technical aspects like deep learning, fast.ai has shown that people with 1-year of coding experience can become world-class deep learning practitioners; you don’t need to hire Stanford PhDs. "
I was in their recent course. No previous AI experience, although I've been coding for 20 years. Using what I learned in the class, I am regularly able to finish in the top 20% on most Kaggle competitions.
Just being able to solve some of those problems allows me to provide employers with major value.
Why would you need more than a year of coding experience to learn deep learning? Isn't it mostly math? I wouldn't think that mastering advanced Python syntax or whatever would be the limiting factor for most people.
I don't know how that could mean anything other than becoming familiar with a deep learning library/platform product and implementing feature requests with it.
In that case, I can readily see that. Especially if they have good teachers and teach the concepts thoroughly.
I don't think that's too outrageous as long as we don't read "practitioners" and hear "experts". Agree that the article is a thinly disguised recruiting advertisment though.
I agree a lot with #3 but disagree heavily with #4.
There are canonical examples of PhD theses that became successful companies, not to mention the dozens of companies who exited (e.g. CV companies to Qualcomm).
The vast majority of PhD thesis based products fail, and they fail because the thesis solves a deep and narrow technical problem, and not a broad market problem.
The examples of successes is both anecdotal and survivor bias, you have to look at all attempts and weigh all the successes against all the failures.
There will be future successful businesses based on theses, among the many more failures. The advice here is really aimed at people who think that a great technical solution to something automatically makes it marketable, which includes many thesis authors.
Someone who's done their real market due diligence, and knows which thesis to productize or which problem to solve, they have a good business idea. But for most PhDs, the advice is good advice -- don't sink all your time & money into a business based on your idea if you haven't done as much study of the market as you have on your thesis topic.
Yeah, Akamai, Google, to some extent Bose come to mind.
Most PhD theses would NOT become companies, so this is good advice for a PhD student but not for a founder. If a founder finds a PhD thesis that is worth productionizing, then the fact that it was a PhD thesis is irrelevant.
The advice is "Don't productionize your PhD" thesis," so I think it is in fact aimed at PhD students in the exact spirit you pointed out, not founders surveying the academic field.
Jawbone founder Hosain Rahman developed his microphone tech while still a Stanford EE I believe. There are also the Quantum Computing startups to spin out of Yale CS. Gene editing pioneered at MIT Broad Institute. And on and on.
Regarding AI / ML as a service specifically. Many startups may find themselves facing the same issues plaguing IBM Watson. Deep learning considers trillion dimensional spaces. Most enterprise prediction problems are simply not of an astrophysical scale. And where domain specific expertise is required. Such as Google Maps use of CV in rendering highres satellite image data to SVG for the browser. Simply having grad student level familiarity with OpenCV might not be sufficient.
A Year of Google & Apple Maps (warning: bandwidth & cpu intensive)
For what it's worth, trends #4 and #5 are both qualified with "AI Specific" further down. There will be some PhD theses that do solve real problems, but many are not going to make a startup successful by themselves.
The family relationship has always been a horrible way of describing the employee/employer relationship. There's divorce or disowning someone, but do you really want to try and draw that comparison when you fire someone? A team has players and they're always being traded to different teams. That's both more positive and honest when you're trying to describe a high-performant culture.
And there's also no concept of unconditional love in a company. There's respect, which should be a given. But you shouldn't be expected to unconditionally love your co-workers.
Negative trends 4 and 5 are not, contra the article, AI-specific; AI just happens to be the intellectual problem of the moment.
The team I was on waaaay back in the prehistory of the web era was warned against hiring too many PhDs. We disregarded the advice. (These were smart guys, yo! They had PhDs!) We spent a couple of years tackling interesting problems instead of relevant problems, and just about entirely missed out on the first web bubble as a result.
I love smart academic people. I love those big brains. I'll never hire one ever again. Give me someone who's more interested in working around a problem than in studying it, and we'll get things done. Give me a PhD and we'll have a nice paper about it about 5 years after anyone other than their thesis advisor cares.
First of all fast.ai is awesome and most of the points they make align quite well with my personal views which of course doesn't mean they are the right views ;) but...
"""Hiring a bunch of academic researchers will not improve your product and harms your company by diverting so many resources (unless your goal is an aquihire)."""
Disagree. Ceteris paribus a previous academic researcher isn't necessarily a worse hire than a graduate or someone who has worked at other tech companies.
I'm really not a fan of this "don't hire people from academia" narrative. A good scientist is well versed in systematic problem solving which doesn't sound like a horrible skillset for someone working at a startup.
Sure there's enough people with PhDs that can't write working code but there's also enough people that can. I'm not sure about the US but hiring a PhD isn't necessarily more expensive than hiring someone with job experience in the tech sector either.
I get their basic idea but it's news to me that even AI startups are on "only PhDs all the time" hiring frenzies.
Without hyperbole, one of the biggest trends to avoid is overgeneralization.
A large part of one's experience at a startup is unique to that context.
Certainly not all. But certainly a lot -- Probably the majority of what counts as advice.
Being able to filter what's generalizable and what's not is the most useful skill an advisor can develop -- ironically, even more than the experience. (And I'm breaking my own hypothesis by giving advice there).
> Then I realized that most of the startups were indistinguishable from one another: nearly everyone was following the same destructive trends which are bad for employees and bad for products
This is rich coming from the founder of a startup that's indistinguishable from all the other coding bootcamp, MooC startups out there with a dash Deep Learning thrown in to make it extra trendy.
As for the actual advice, I think it's pretty hit or miss.
> 1. Venture Capital
Nothing in her description of VC is wrong here. But the lesson: "VC is to be avoided" is an oversimplification. VC money is a tool. It's not always the right tool, and it certainly has its downsides. But it has its upsides too. In many cases it's the only way to get a business off the ground.
> 2. Hypergrowth
I don't really know where the author puts the line between hypergrowth and just growth, as PG has said: starts are growth. They must always be growing, is there such thing as too much growth? Maybe, if really bad things are happening in the company because of the growth like people are burning out and quitting... but really that's only bad because eventually it will hurt growth. So I think the simple lesson is grow as fast as possible, don't be short sighted.
> 3. Trying to be “like a family”
This one I agree with. Trying to be like a family in a professional setting is dishonest, and eventually the truth of the situation becomes clear. It's better to be honest from the start.
> 4. Attempting to productionize a PhD thesis
We'll have to tell Larry Page he's wasting his time trying to turn that dumb Page Rank thesis into a company. There are a lot of companies that fail this way and there are a lot of PhD theses that shouldn't be commercialized but people try none the less. As with all things startups, there's a thousand failures for every success and the only way to tell the difference is to actually found the company.
> 5. Hiring a bunch of academic researchers
The only company I've really observed doing this is Google and it seems to have worked out for them. I can easily believe that their are companies out there that higher a bunch of researchers to do a job they're not really capable of doing. I've also known academic researchers who were able to have a huge impact in the role they were hired for even though it wasn't really research. The general statement of this is that you needed to hire people for roles they want to and can perform. Researchers might be a particular anti-pattern in this but it's far from the only one.
Is doubling in a year even "hypergrowth"? Doubling revenue every a month or two? Sure. Doubling revenue in a year is a very strong growth rate for an established company, but it doesn't strike me as particularly hypergrowth.
Doubling headcount in a year seems also fairly routine for early-stage companies (assuming funding and cachet is in place).
#3. This is so truth. People have no idea what they talking about when they say "we are a family". Forget this whole family crap! Hire skilled and talented people. Let them do their job. Give them freedom and don't stand in their way. Employees barely know what a friendship is, let alone a family.
> You and your adviser picked your thesis topic because it’s an interesting technical problem with good opportunities to publish, not because it has a large opportunity for impact in an underserved market with few barriers to entry.
This is obvious, yet an easy mistake. Behind it is applying the same values to a business as to a PhD - not only in topic selection, but also in execution. It's related to the conundrum that solving a problem that real people have now does not always require a technically great soluton.
Even a revolutionary business does not necessarily require a technically great solution.
Academic success requires academic values and skills; business success requires business values and skills.
Of course, if you want to create a technologically revolutionary business, you'll need both.
A trend that I would like to see more startups buck is the use of the independent contractor model. While I believe startups like Uber and Lyft could not achieve the scale that they have w.o this model, there are smaller startups & markets where I believe the increase in value to customer service can really differentiate a product.
Studies have shown that customer service, ability to execute & perform, and general quality of service correlate with whether or not the employee is an IC/temp or not. I've wondered if any startups have gone the opposite direction and hired folks FT/PT for roles that otherwise would seem to be filled by ICs.
Adding pub-sub/queue/message broker in your app when a normal when a normal connection to SQL db from your web app is more than sufficient. There are a lot of companies that have kafka in their stack when they actually don't need it. This is one thing i have seen a lot.
On front-end side adding redux when a normal react-app would be just fine.
"even for highly technical aspects like deep learning, fast.ai has shown that people with 1-year of coding experience can become world-class deep learning practitioners"
Yes, any Joe can train deep network with dozen lines of Keras. Sure, your startup can use off-the self models and tweak it a bit... That doesn't make you "world-class deep learning practitioner". If you are designing the network for new problem domain, there are thousands of decisions to make - everything from hyperparameters to network architecture to distributions in data. Making these decisions without having developed lots of intuition and good foundations is very hard. I often find usual developers without ML background and experience completely lost in these cases. Debugging a model that doesn't work is super hard. There are no IDEs, no breakpoints, no watches and in fact not even error messages. Its purely statistical debugging and probabilistic fixes.