Impactful storytelling

  • Make clear end of your journey for others

  • People do not remember facts, they remember stories

    • Make user THE hero, the center of story

    • Make your vision a story

  • If you use influencers invite them to debugging and tweaking

  • Venture capitalists want transformation, not disruption or revolution

    • How is my product going to change their life?
  • If highly technical, make parallels to mundane events

  • In B2B

    • Find out what does fuel the buyer ‒ they should see a change

Adventures in building a hardware startup

  • Make something people want

  • Hire people better than you

  • There is very little common between prototype and mass production

  • Known known is good
    • Known unknown ok
      • Unknown unknown will kill your startup
  • Noone can deliver what we promise
Source: Kickstarter
  • How to avoid delusion of Kickstarter?

    • Ship quickly, get it to the people in a small amount

    • “Selling” prototypes

      • Barely works

      • Offer a few on Kickstarter (100-1000 pieces)

      • Prototype should be big and ugly

    • Get an Advisory board

      • Offer 1 %

      • They have the knowledge you don’t have

    • Mass production

      • Hardware club ‒ help for startups

      • Dragon innovations

      • Start designing with the factory

      • It takes approx 1 year to make the HW

Creative problem solving using server less and ML

  • Microsoft has cool Azure automated IMG classificators for non-ML specialists

  • Basically a 1 hour Microsoft Azure promotion with some of the most basic NN classification information

Interview with Jakub Nesetril, VP Product Development at Oracle

  • Do not outsource, but use services ‒ they save time&energy and tent to be way deeper than your implementation


  • Do less ‒ type less, think less, do more ‒ whole presentation

  • Setting up a server is hard: slicing, load balancing, latency…

  • Solution? Use Zeit ‒ server pushing in one command

    1. Builds from package.json (Node.js) or Dockerfile
    2. Free minimalistic version
    3. After :now: command you get a new link
    4. Once the build is completed, you can directly access the build via given link
    5. Then it can be pushed as main version
  • Look more into Next.js

    1. Embrace the filesystem
    2. Embrace conventions
    3. Make it lightweight and flexible, obsessively
    4. One command to develop: next
    5. One command to build: next build
    6. One command to export statically: next export
    7. One command to server-render: next start

People & algorithms: Build AI-driven features that don’t cause harm

  • Many problems

    • ML models get things wrong and we can’t always explain the results

    • Underlying bias problem in the data

    • Data is not inherently objective or fair

    • Removing bias requires human intervention

  • Examples

    • Accuracy at face recognition white × black faces

    • FB showing year review of death in family with balloons around it

  • Design can make intelligent features seem less scary and more worthy of trust

    • Google search ‒ tries to look like “machine, human untouched” feed of truth
  • Ask the (uncomfortable) questions

    • Ask questions about data sources

    • Design methods for feedback

  • More at AI for Good

Power of programming

  • The best way of helping others, the world is not by working for nonprofit, not giving money, but

    • Becoming a software developer!
  • Developers break the rule of supply and demand

    • Our work multiplies, it costs the same to make code for 5 or 5 milion People

    • WhatsApp was initially created by only 40 people yet it improved live to milions

Bots on the net: The good, the bad, and the future

  • Bots are going to be a multibilion industry

  • Bussines see savings × customers do not trust them = gap

    • What if people couldn’t distinguish between human and bot = Turing test

    • 1st to pass the test for some people in their time (1966)

  • Specialized AI or Microworlds

    • Help you buy a shirt, fix a computer or get weather information

    • Work well in their space, but can’t expand

    • Don’t need to learn

    • State-of-the-art for most problems

  • Encoding behavior exposes bias

    • Latent bias ‒ build-in

    • Selection bias ‒ see People & algorithms: Build AI-driven features that don’t cause harm

    • Interaction Bias ‒ acquired when the bot communicate with us and reflect it back

Source: Google Images
  • Loss aversion ‒ people need to gain 150 $ after lost 100 $ to feel even

  • Statistical learning ‒ there is roof to how much data increases the accuracy

  • Algorithms drive ML chatbots; Big Data is the fuel

  • What to do?

    1. Self identifying rules: “I am a bot.”
    2. Licencing: A licence to operate a certain bot

The age of Assistance

  • The web changed the world 20 years ago
  • Mobile-first changed the world in less than 10 years ago
    • Getting internet to even more people ‒ 5 bilion by now
  • AI-first!
    • Assistants are really good at accessing directly really deep points in other apps
      • Specific place in Street view; translating sentences; setting an alarm;
    • Personalization
      • Remembering context, favorites, work, home; asking for personal info gathered from emails; finding pictures
    • 3rd party integrations ‒ for near future
    • Going to be a revolution, because AI saves time for users