In Timing Technology: Lessons From The Media Lab, Gwern highlights an interesting phenomenon. Many visionaries demonstrate remarkable prescience about what a future transformative technology will look like (e.g Mobile Phones, VR etc), but are seemingly unable to capitalize on this foresight.

This inability stems from their failure to accurately predict when a transformative technology will happen. The return on investment from startups follows a power law, meaning if you don't back exactly the right unicorn horse, you miss out on ~all the gains. The power law explains a common phenomenon seen in Venture Capital, termed "stampeding". When it becomes clear that a sector is experiencing explosive growth, both entrepreneurs and VCs will flock to it, and only a few lucky ones will capture all of the value. This stampeding effect is a well known, often derided characteristic of venture. However, there is a second, less often discussed failure mode, namely leaving a billion dollar note on the floor.

Starving a transformative technology of investment past the point of possibility means VCs are leaving value on the table. The paragon of this failure mode is Oculus VR. Palmer Luckey was 18 years old, had no VC funding, and lived in a camper van on his parents drive when he developed the first Oculus prototype. Oculus went on to be sold to Facebook for $2.3B. How was it possible for a unicorn to be created by a teenager with no industry experience or capital? I believe it was Orthogonal Technology Development.

Developments in the consumer mobile industry (Battery capacity, Display resolution etc) enabled consumer VR to be created for an acceptable price. Luckey had no capability of doing this R&D himself, he simply struck at the opportune moment. The amount of funding available to early stage startups precludes any hope of doing significant fundamental R&D on Input Technologies. This means that successful startups must wait for developments in an orthogonal field to enable new capabilities for them to capitalize on. A venture fund tracking fundamental input technologies, maintaining a database of failures from the past, and understanding government R&D (typically the earliest exploratory R&D), will have an enormous advantage.

The utility of this information differs for a VC and an entrepreneur. Unlike a VC fund, an entrepreneur does not have the ability to distribute their bets among multiple startups, they either start a company in a sector or they do not. Whilst an entrepreneur can make use of this information to decide if their venture has a possibility of success, the inability to hedge across multiple ventures in a sector reduces the utility.

Good startup recipe

...a good idea will draw overly-optimistic researchers or entrepreneurs to it like moths to a flame: all get immolated but the one with the dumb luck to kiss the flame at the perfect instant, who then wins everything, at which point everyone can see that the optimal time is past.
— Gwern

There are some interesting commonalities among good startup ideas:

1. Due to the Babe Ruth Effect, you need the TAM to be large enough to be the fund leader. Determining the TAM is challenging, as truly transformative technologies create new markets (i.e Uber reduces friction of getting a taxi, so demand increases).
2. All the best ideas have been tried before, if your idea has not been tried before by the government or industry, it's almost certainly not a good idea. Multiple discovery is the norm, not the exception.
3. Given that it's been tried before, the startup needs to have failed due to uncontrollable external factors (e.g Display Resolution, Battery Technology, Macroeconomic conditions etc).
4. The landscape needs to have changed to make the external factors causing the prior failure no longer valid. Investment in an orthogonal field must have reduced the price / improved the capabilities of prerequisite technologies.

A systematic database of ideas like this gives VC funds a starting point when evaluating potential unicorns.

Case Studies

We've determined that understanding Input Technologies and obviously good ideas (bottom up vs top down) is a useful exercise for both VCs and entrepreneurs. We shall apply this theory to prior ventures to further validate the approach.

Pets.com

Let's begin with the canonical example of 90s hubris - Pets.com. The founders of Pets.com were thwarted by 2 key things: number of people on the internet, and the proportion of those comfortable with transacting over the internet. In 1999 only 113M people were using the internet, and a unquantifiable but certainly smaller fraction of those were comfortable paying for products online. Does this mean that selling pet specific products over the internet is a bad idea? Luckily the universe has run the experiment for us. Chewy.com is exactly the same idea, only in 2010 when 288M people were using the internet, and they'd all paid for things online. Fundamentally, there was nothing the founders could have done to change this, save for a parallel venture in internet infrastructure or online payments.

SpaceX

What was once the realm of governments became the realm of a private company. SpaceX has done this twice now, both with launch capabilities and with Starlink. Starlink is the primary revenue driver of SpaceX, but was attempted before by Motorola with their Iridium constellation. It does not require Nostradamus to know that global satellite internet is a desirable thing. Mapping and understanding when the prerequisite technologies make it commercially viable (e.g reusable launch vehicles reducing $/ton to orbit by OOMs) is key.

The creation of a transformative startup is a virtuous act, as it itself is a prerequisite for the next generation. What startups are enabled by $/ton to orbit dropping by another OOM with Starship?

Tesla

An all electric car is not a new idea. Below is a Popular Science magazine cover from September 1975.

So why did it take 40 years to get an electric car people actually wanted to buy? JB Straubel realised that the time might be ripe for the first consumer electric car when he was helping out with the Stanford Solar Car Challenge.

Straubel and the solar team kept fixating on one topic. They realized that lithium ion batteries ... had gotten much better than most people realized. Many consumer electronics devices like laptops were running so-called 18650 lithium ion batteries. ... "We wondered what would happen if you put ten thousand of the battery cells together... "We did the math and figured you could go almost one thousand miles. It was totally nerdy shit ... but the idea really stuck with me.

Elon Musk Biography — Ashlee Vance

This realization went on to be codified in "The Secret Tesla Motors Master Plan (just between you and me)", which is truly one of my favourite documents of all time. If you do not have this level of clarity on why for fundamental reasons your startup is going to win, I don't believe you can have the conviction to push through the inevitable dark times. Treat that document like the Constitution, refer to it in times of hardship.

Netflix

Although beginning as a mail order DVD service, the founders of Netflix one day envisioned distributing media over the internet, hence the name. The fundamental prerequisite for this was bandwidth (both cost and quantity). If Netflix started distributing video over the net too soon, their bandwidth costs would burn through all their capital. Netflix launched WatchNow in January 2007, when the average global internet speed was ~2.5Mbps. This just so happens to be the minimum viable speed for streaming 720p video. Funny that.

By 2011, Netflix had become by far the largest user of peak bandwidth on the Internet. Understanding within a single digit year range when it would be possible for video media to be distributed over the net at palatable costs would have made identifying investment opportunities in this space much lower risk.

The Checklist

Humans are very error prone, which is why in high stake environments checklists are fundamental. In his post, Technology Forecasting: The Garden of Forking Paths, Gwern outlines a Munger-esque checklist for technology forecasting:

1. Are there any hard constraints from powerful theories like thermodynamics, evolution, or economics bounding the system as a whole?
2. Breaking down by functionality, how many different ways are there to implement each logical component? Is any step a serious bottleneck with no apparent way to circumvent, redefine, or solve with brute force?
3. Do its past improvements describe any linear or exponential trend? What happens if you extrapolate it 20 years (in the grand scheme of things, a minor error of timing)? If the trend is exponential, is it a set of stacked sigmoids indicating considerable flexibility in choice of technology/method?
4. What experience curves could this benefit from in each of its components and as a whole? If the cost decreases 5% for every doubling of cumulative production, what happens and how much demand do the price drops induce? Are any of the inputs cumulative, like dataset size (eg. number of genomes sequenced to date)?
5. Does it benefit from various forms of Moore’s laws for CPUs, RAM, HDD, or GPU FLOPs? Can fixed-cost hardware or humans be replaced by software, Internet, or statistics? Is any part parallelizable?
6. What calibre of intellect has been applied to optimization? What amount of ‘constant factor’ slack is left in the system to be micro-optimized away?

Given humans are fallible, a systematic application of checklists like this (augmented by AI) can further derisk investments.

Designing a systematic VC fund

One idea is that a VC firm could explicitly track ideas that seem great but have had several failed startups, and try to schedule additional investments at ever greater intervals (similar to DS-PRL), which bounds losses (if the idea turns out to be truly a bad idea after all) but ensures eventual success (if a good one).
— Gwern

Although, "ideas that seem great" will be steeped in hindsight bias, I believe that an attempt should be made to structure a platform for VC funds / entrepreneurs that is designed as follows:

1. Explicitly tracks developments in Input Technologies, like batteries (Wh/kg, Wh/$), bandwidth (Average Mbps per household), computation (TFLOP/W, TFLOP/$), launch capacity ($/kg to orbit), sequencing cost ($/Mb of DNA).
2. Maintains a historical record of previous startups attempting to create an obviously good startup idea (e.g humanoid robotics, pizza delivery by app etc).
3. Using the above 2, build a set of trigger conditions for obviously good ideas (e.g 150 Wh/kg + 100$/Wh for electric cars). Then, when the opportune window arises, the fund can back multiple entrants or inculcate an entrepreneur with the idea.

Despite difficulties in data sourcing, data quality, hindsight bias and others, Kairos will be an attempt to systematize these ideas. Being aware of the "technical window of opportunity" does not supplant VC understanding of the broader market, it only endeavours to derisk investments that are fundamentally too early.