Software Programs the World

I got a lot out of this podcast! I initially listened to it driving, but the insights were so interesting that I had to listen to it twice afterwards, taking notes to absorb it all. It's an exciting brave new world we have coming.

The most interesting parts:

  • Foundations (jump to clip)
    • Foundational element one: Moore’s Law has "flipped" over last 7-8 year
      • Traditional Moore's Law: a new chip was release every 1.5 yrs that was 2x as fast at same price (this lasted for about 40-50 years)
      • This is the dynamic that drove mainframes, minicomputers, PCs, and smartphones
      • About 7-10 years ago, this ended; chips "topped out" at about 3 GHz
      • Now the dynamic has "flipped" so that every 1.5 years, chips are just as fast but half the price
      • This is a "massive deflationary force ... where computing is becoming essentially free"
      • In this business, we “chart out the graphs and assume we get to the end state," one where chips will be essentially free
      • Chips will be embedded in everything (this is a new world)
    • Foundational element two: all of those chips will be on the network
      • Wifi, mobile, wired, etc.
    • Foundational element three (continuation of "Software Is Eating The World"): software, then, will allow you to program the world
      • Cars, things in the sky, buildings, homes, businesses, factories, etc.
      • "This is just starting"
      • Entrepreneurs are more interesting, more aggressive than ever before because they assume "if there is something to be done in the world, software can be written to do it"
      • Consequences: "...investing in markets that 7-8 yrs ago we would have never anticipated operating"
  • New platforms (jump to clip)
    • Platforms will be different than what we've had until 5-10 years ago: the platform was a new chip (faster) and a new OS
    • Platforms today: distributed systems, scale out systems
    • These are not on a chip, rather built across a lot of chips (distributed systems)
    • Cloud was first example (AWS) - can now create a program that can run across 20k computers (run for 1h, cost $50)
    • Rise of Hadoop, Spark (distributed processing)
    • Financial technology: bitcoin, cryptocurrency
    • Now: AI (machine learning, deep learning) which is "inherently parallelizable" - can run across many chips and get very powerful as they do so
    • "Can do things in AI with distributed computing that you couldn’t imagine 5y ago"
  • The GPU
    • Initially developed for gaming for very high resolution graphical processing → unexpected uses
    • "New application of an old idea"
    • Thirty years ag in physics lab -- if you need a simulatio with large number of parallel calculations (e.g., black holes, biological simulations), write algorithms to parcel problem into pieces and run in parallel
    • In this days days: vector processors ("sidecar computers")
    • 30y later: GPU is basically a vector processor (sits alongside CPU)
    • Ben Horowitz at Silicon Graphics → physics applications, flight simulations, computational fluid dynamics
    • Simulating real world → need same capability (exact same processor)
    • HW platform emerging: NVIDIA
    • NVIDIA has become “seemingly overnight” → market leader in GPUs and chips for AI
    • All entrepreneurs in AI building on NVIDIA chips (in contrast to Intel in previous years)
  • In AI, what are the things that lend themselves well to startups versus larger companies (e.g., FB, Google, Apple)
    • Challenge: people think of AI as narrower than it really is; rather, it is an entirely new way to write a computer program (broadly applicable to problems)
    • Could use AI to analyze consumer data (hard to compete with Google)
    • BUT: many areas where no one has any data yet (HC, autonomy)
    • Big company advantage: lots of data
    • In reality: just a new way to write a program
  • Interfaces
    • The smartphone was an advance over WIMP
      • WIMP interface: windows, icons, menus, printer
    • That was an advance over text based interface of DOS
    • BUT -- life is different
    • Natural interfaces: natural language
    • AI can enable natural language and natural gestures
    • Opportunity to build interfaces for things you couldn’t before
    • One idea: what applications couldn’t you have before because there wasn’t a workable use interface for it
    • Amazon/Alexa
    • Not tied to old generations
    • No “strategy tax”
  • Ability to leapfrog 
    • For many major new advances, interfaces depend on platform
    • But - big companies also have strategy tax -- existing agenda, have to fit next thing into old platform
    • Example: Amazon has taken lead from Apple, Google -- even though it flopped with phone!
    • Lack of phone became an advantage! Clean breakthrough product
  • But where can startups play?
    • A year ago, would have said AI would be domain of big companies (can afford engineers, hardware; data sets)
    • All three have changed
    • AI technology is standardizing (open source → cloud)
    • AI as a service
    • “AWS for AI” (Google, Amazon, Microsoft, etc.)
  • TensorFlow ("This is a big deal")
    • A lot of students on TensorFlow (“trickling down very fast”)
    • Most teams at hackathon had AI and machine learning components
    • Hardware costs coming down across the board
    • In one year -- AI supercomputing chips with algorithms in the cloud (massive deflation)
    • Big data sets -- startups can assemble big data sets BUT...
    • Newest generation of experts -- focusing on small data sets
      • "They'll say, Primitive and crude machine learning required large data sets but not the newer algorithms (they can work on small data sets) - early but enticing (brings problems into small company realm)
    • With these GPUs -- can create simulated versions of the real world using video game tools (can train AI)
    • Earthquakes, floods, thunderstorms, swarms of birds
    • Train AI - "AI actually has no idea it’s working in a simulated world and not the real world"
    • Potentially: run millions of hours of simulated training at very low cost
    • Google’s Deep Mind data set -- "game playing itself"
  • Why are simulations so important?
    • Ten years ago - AI, neural nets, and deep learning were frowned upon
    • Improbable (company) - can do large-scale scale out simulations using cloud computing technologies and new, proprietary technology
    • Can get a complete picture of the world, can generate a data set
    • "It’s expensive to make things happen in the world (physical changes...building roads, planes...are hard, expensive; have consequences)..."
    • In contrast: simulation, run experiment, introduce change -- easier, cheaper, no consequences
    • Can run millions, billions of simulations
    • Can make real world decisions with more foreknowledge
  • What are other areas where you can think of real world applications of this technology? (jump to clip)
    • New platforms: health + computer science
    • AI hardware for different type of programming; today, Google has a new chip for deep learning cloud
    • New breakthroughs for quantum computing (more powerful deep learning systems)
    • Chip in everything: platforms to run/manage those chips
  • Theme: tech reaching into new places; “tech is outgrowing the tech industry”
    • Thesis: software is eating the world BUT “hard investments” (Soylent, Oculus, Nutribox)
    • Oculus was actually software (breakthrough tech often needs new hardware)
    • Soylent and Nutribox -- same thing
    • Big believers: big breakthroughs in knowledge (Turing, Shannon) -- new model of the world, companies that build on that new knowledge
  • SaaS -- acquisitions; what is left to do there? SFDC or vertical or totally new platforms
    • SaaS as old versions of things in the cloud (WDAY, SFDC, SFSF) -- big categories
    • Changed from on-premise to cloud: seeings sw applications for things that in the old days were cost prohibitive (“screwing it in and hiring army of Accenture consultants) (e.g., expense reporting: Concur) -- new things come into economic viability
    • What was unviable before?
    • Can also scale down to small companies as buyers (<1k employees) - Oracle Financials v. NetSuite
    • Verticals (real estate, construction)
    • Interesting trend
    • Historically: SAP, IBM, Oracle … accessible to top 500-1,000 companies in a handful countries
    • So previously big companies in big countries had an advantage (dominated by 2k-3k multinationals globally)
    • N. Am and Western Europe v ROW
    • Interesting conclusion: smaller company or not one in the Western world (leapfrog similar to what happened with telecom)
    • Larger companies may have harder time adapting
    • Maybe: power shift from larger to smaller companies
    • Companies in western world to those in ROW
    • “Leveling of playing field”
  • The macro view of the economy -- “world is starved for innovation and growth”
    • $10t of capital held in gov’t bonds trading at negative yields
    • “Paying the bank interest”
    • “People cannot find enough productive places to put their capital”
    • Negative, conventional view: starved for growth
    • Positive view: $10t waiting for new opportunities (HC, education, consumer products, media, art, science, cars, housing, etc.)
    • “What needs to be done in the world?”
    • “World has never been more ripe for a VERY large wave of innovation that would be quite easy to finance”
    • More money than ideas and creative, effective people
  • Company building and founders -- types of founders; what has changed?
    • “Gotten more risk tolerant”
    • “We’re much more interested in the magnitude of the strengths than the number of weaknesses”
    • Lack of experience is a strength: “Hard to rewrite the world if you’re too steeped in the world”
    • Financial terms: “buying volatility”
    • “World class strengths where we care about them”
  • One piece of advice
    • Management: “The most common mistake founders is making decisions based on very proximate perspectives without taking the time to think about how others in the company will see the decision...let’s look past the person I’m talking to.”
    • Strategic: “People need to raise prices.”
    • Most companies have sophisticated views on product, design, engineering and naive views on prosecuting a campaign
    • One dimensional view between price and volume (“pricing cheap, selling more”) 
    • Two dimensional view
    • Raise prices, and you can afford a bigger sales/marketing effort
    • Most companies have prices that are too low to get people to buy
    • Too hungry to eat problem
    • Vicious cycle
    • When you charge higher prices, people take the product more seriously, impute more value, make a serious decision, and when they buy it, they experience a greater sense of engagement, commitment, and stickiness