Profile cover photo
Profile photo
Mark Cummins
1,369 followers
1,369 followers
About
Mark's posts

Post has attachment
The energy expansions of evolution. An overview of the history of life in terms of the available energy sources.

Post has attachment
Some interesting data on startup survival and exits:

--Startup survival decreases almost exactly exponentially (in units of time measured by funding rounds)
--16% of startups ultimately get acquired (plus a tiny number IPO)
--Most start-ups get acquired early in their lifecycle (prior to Series B)

Very interesting to see this data. The 16% acquisition percentage seems spot-on (most investors will say 1-in-10, which is close).

One thing that surprised me was that the failure rate was so constant with time. My impression was that there was a bottleneck at Series A, with a higher failure rate for seed-stage startups, and a better survival rate after Series A. You can see a small effect like this in the graph, but it's not very pronounced. The data here comes from Crunchbase, which is not perfectly clean, so it's possible the explanation is that that early stage startups are not all being captured by Crunchbase, but later ones are. Or it may be that there really is no bottleneck!

Anyway, it seems a good rule of thumb is that 50% of startups that make it to one funding round will not make it to the next (whether because of failure, exit or becoming self-sustaining). Also, about 15% of all startups exit, and 80% of exits happen before Series B.

Post has shared content
Best thing I have read in ages! Some novel ideas on the Fermi paradox. The basic idea is that advanced civilizations may choose to "sleep" for a few hundred billion years until the universe is much colder, which allows you to do a lot more computation with a fixed amount of resources. This is a beautiful idea, and it makes some predictions about what you might currently observe if it were true. Even if the Fermi paradox has another answer (e.g. we are first for some reason), this is still a potential future for us.

There is also an informal FAQ here:
http://aleph.se/andart2/space/the-aestivation-hypothesis-popular-outline-and-faq/

Post has shared content
Another Steve Yegge rant, which got me to look at Kotlin, which appears to be a simplified Java with some of the same syntax that people like about Python.

http://steve-yegge.blogspot.com/2017/05/why-kotlin-is-better-than-whatever-dumb.html

Direct link if you want to try Kotlin yourself:

https://try.kotlinlang.org

Post has shared content
I'm very excited about the work many people at Google have been doing to develop our own custom systems for accelerating training machine learning models. Version 2 of our Tensor Processing Unit is a device that offers 180 teraflops of compute, and is also designed to be connected together with an ultra-high-speed network into a TPU Pod of 64 of these devices, offering 11.5 petaflops of compute. These systems are programmed using TensorFlow.

These TPU devices will be coming to Google Cloud as Cloud VMs with TPUs.

Post has shared content

Post has shared content

Post has shared content
+Julian Ibarz on interpreting the world from Street View data. The Google Geo teams really have the most interesting problems and datasets to work on in computer vision today.

Post has attachment
Very plausible scenario. Amazon is much better placed than Google to make money out of voice interfaces. Google may finally have to find something beyond ads to survive.

Post has shared content
Thanks to a laptop charger rescue by John I was able to finish up the night sky pano of Shark Fin Cove!
Photo
Wait while more posts are being loaded