Open Office Hours 2023-05-30

2023-05-30 16:00~17:00 yodayoda

Israel: impact on studying abroad and starting a business

• Startups: the highest number per capita in the world.” Startup Naion”.

• Spin-out from the military: The military plays a talent discovery, hands-on training, and incubation function. Military connections are strong and ideal for finding entrepreneurial friends.

• Strong security

• 8200部隊があるからですかね？The ←8200 unit is only part of the story, but yes.

  • What is the Israel Defense Forces’ “Unit 8200” that creates cyber companies? | Forbes JAPAN Official Website
  • Unit 8200 - Wikipedia

• military affairs A realization that there are seeds of radical startups in technology. Searched for.

• [How the Tomahawk Missile “Pinpoint Bombing” is performed.

  • The CEP (Circular Error Probability) is raised by matching satellite data acquired in advance with observation information acquired by Tomahawk’s own onboard cameras and sensors.
  • The initial idea was that this could be applied to drone and used for “pinpoint delivery” of buildings, etc.
    • GPS accuracy should not be enough, and vertical directional information is also expected to be necessary for a practical drone society.
    • Thesis Statement.

• Pinpoint ≒ If you can get a point, you can draw a line.

  • It could be used to design drone flight routes.
  • It is the 3D data that is acquired both before and now.
  • The business was set up as “map for automatic drone flight” and started. Funding happened to go well.
  • However, the spread of drones has been much slower than expected. The direction of the project has changed as there is almost no use expected in urban areas.

• Pivot to “Map for self-driving cars.”

  • Considering the market, expansion to the U.S. is a must. The company was reorganized and became a U.S. corporation headquartered in San Francisco.
  • More failures. Two causes of defeat.
  1. Building a human network in the U.S. (sales force) is a hurdle. - You can’t even make a sales appointment without going into the inner circle.
     • Need to build up SOC in the US
  2. technical capabilities. It attracts some of the strongest people in the US. A start-up guide who has lived in the area for a long time says to me, “What you are trying to do is like watching a Super Bowl (football) game and asking if you can be a part of it. I can hear the sound of your spine breaking”. It wasn’t far off.
     • Waymo Waymo - Wikipedia

       • A self-driving car development company under Alphabet, created on December 13, 2016 through the spin-off of Google’s self-driving car development division.

     • The core members are from the Google Maps team. Maps are extremely important in automated driving technology.

• Currently moving to Street View service for humans

OpenStreetMap : Mapbox : GoogleMap Why Mapbox was able to become a unicorn

• GoogleMap API is expensive.
• Diversification of map design
• Google is not trying to make it easy for individual companies to use.
• Maybe there’s a tendency in Google to focus on big commerce because it’s hard to be valued for doing things that meet the diverse needs of individual small companies.
  • Google is truncating the long tail, so there are opportunities buried in the truncated long tail for other companies to become unicorns.
• Like GCP, Google is basically a “lordly” business and is not customer oriented.

Will we move from linguistic data to environmental data in the long term?

• Currently, LLM has evolved using linguistic data only because linguistic data is “easy” to collect
• Training data for image generation AI and data for text-to-image correspondence were needed.
  • The base is data on img tags and their ALT texts collected by crawling the Internet
  • Frankly, it’s a pile of garbage, but it still got to the point where Stable Diffusion could generate images.
  • Then, after seeing that release, NovelAI did additional training on data from Danbouru, “a site where humans are tagging illustrations with beans,” and the “subjective ratings of fans of illustrations” regarding the generation of illustrations jumped!
• Take the easy data first, and eventually you will need “harder data”.
  • We can’t predict when that demand will jump.

Where does the Mapbox data come from?

• OSM-based
• It’s just a bunch of data, no usability or design.
• After all, neither OSM nor GoogleMap were taking a close look at unfulfilled customers to identify “what customers really want”!
• What about Mapillary?
  • private-sector business
  • Users contribute data by allowing users to create StreetViews.
  • Acquired by Meta.

Operational Know-How

• When data is needed, there is a big difference in the cost of data between “entities that only do data acquisition” and “entities that can acquire data while doing business”.
• Using a large number of people to collect data is a bottleneck in the education of that large number of people.
  • Operational knowledge of how to structure and educate people to collect data without expertise will be valuable.
• Being able to brush up on that knowledge while doing business in the long term would be a differentiator.
• The strategy of making high-risk, high-return investments in surplus to meet the demand we see now and generate revenue to sustain the organization in the long term makes sense. - Barbell Strategy

Large amounts of data, storage billing a problem?

• That’s as far as we can go…
• You should consider whether the huge video and photo data itself is a differentiating factor
• Danbouru’s data that generated significant impact in the NovelAI case study was “a small amount of, but well-tagged data”.
• Value could be created not in the video or photo data by itself, but in the data that corresponds to it with other information.
• If I were a “super person who can put all the stored data in my brain,” it might be good to think about what correspondence would make each specific job easier!

---

This page is auto-translated from /nishio/公開オフィスアワー2023-05-30 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I’m very happy to spread my thought to non-Japanese readers.