Newsletter 37 – Software 2.0

Published on Author malmLeave a comment

Software 2.0

Andreij Karpathy in an important post entitled software 2.0 suggests that we are witnessing nothing short of a mass scale phase transition in software development to a future in which the job of “programming” will mostly wither away as developers spend their time tweaking and tuning weights in neural networks with pre-built models:

It turns out that a large portion of real-world problems have the property that it is significantly easier to collect the data than to explicitly write the program. A large portion of programmers of tomorrow do not maintain complex software repositories, write intricate programs, or analyze their running times. They collect, clean, manipulate, label, analyze and visualize data that feeds neural networks. … Software 2.0 is not going to replace 1.0 (indeed, a large amount of 1.0 infrastructure is needed for training and inference to “compile” 2.0 code), but it is going to take over increasingly large portions of what Software 1.0 is responsible for today.

And in a somewhat overly breezy denouement, he makes the startling observation that there will be no software development beyond that point as it will be done by the AGI developed over the coming years:

In the long term, the future of Software 2.0 is bright because it is increasingly clear to many that when we develop AGI, it will certainly be written in Software 2.0.

And Software 3.0? That will be entirely up to the AGI.

His friend and collaborator Pete Warden suggests that what Karpathy means is that “deep learning is eating software“, a trend he also views as inevitable:

There will be a long ramp-up as knowledge diffuses through the developer community, but in ten years I predict most software jobs won’t involve programming. As Andrej memorably puts it, “[deep learning] is better than you”!

As if this existential dilemma weren’t enough, software professionals are increasingly seen with suspicion and dislike by society at large.   As The Guardian points out, they have in some senses have replaced bankers in that regard:

Being in tech puts a badge on you. Things are going bad for a large section of the economy in this area and here’s a shiny beacon of people getting paid far too much for what they do. It’s a very easy target especially if you mark yourself as one

Here are some “brilliant assholes” from Dropbox on show in San Francisco illustrating a sense of entitlement that would rub anyone up the wrong way:

In this brave new world of code zero, Ev Williams suggests that tech firms will compete primarily on trust. Good luck with that.

Artificial Intelligence and Machine Learning

Evidence is growing that Amazon are looking to really push into enterprise AI. The Information disclose the presence of an internal AWS initiative codenamed Ironman which appears to target precisely this space and may be announced at re:Invent later in November:

Central to the AI push is a program, code-named “Ironman,” which includes a new AWS cloud “data warehouse” service that collects data from multiple sources within a company and stores it in a central location. AWS launched a data warehouse service five years ago, which rearranges how data is stored to make certain queries faster. Ironman is an AI-specific upgrade that is designed to process massive amounts of data for use in machine learning computing jobs, said two people with knowledge of the program.

The site offers a compilation of pre-trained deep learning models with demos and code.

The latest Boston Dynamics Atlas robot does parkour:

The Slaughterbots spoof video that went the rounds this last few weeks is a parody beyond Black Mirror.  There’s no escape from the uncomfortably plausibility of this scenario though – chronicle of a death foretold:

Streaming Video

Bloomberg on why Apple should buy Netflix.  It’s all about facing up to Amazon:

both companies are well aware that the mightiest competitor they face these days is Amazon. A tie-up creates a competitor that would equal the Jeff Bezos juggernaut.

Meanwhile The Information are reporting that the brutal combination of rising programming costs and falling profits mean that many of the incumbents in the cable TV space will face disruption and even extinction within a decade.


Good post on how to invoke Rust from Python which led me to try doing the same using extern C and cffi.  This recipe is essentially the same for interfacing Python with C++ or C based code.  This Rust code:

pub extern "C" fn mydouble(x: i32) -> i32 {
    x * 2

can be built into a library using cargo build and then invoked from Python as follows:

from cffi import FFI

ffi = FFI()
int mydouble(int);

C = ffi.dlopen("target/debug/libdouble_ffi.dylib")
for i in range(10):
    print("Attempting to double {} = {}".format(i,C.mydouble(i)))

The latest version of spaCy 2.0 the leading OSS  has been released with considerably enhanced support including for languages other than English and updated documentation and APIs.  It was already a compelling proposition and could now overtake the venerable NLTK as the de facto standard Python NLP library:

The new version gets spaCy up to date with the latest deep learning technologies and makes it much easier to run spaCy in scalable cloud computing workflows.


The blackest black is vantablack:


Technology, Culture and Society

Internet connected block of wood display for your smart home:

Fig jam.  Everyone’s favourite topic is themselves because it feels good:

On average, people spend 60 percent of conversations talking about themselves—and this figure jumps to 80 percent when communicating via social media platforms such as Twitter or Facebook.

Eye-opening and remarkably insightful comments from Sean Parker, founding president of Facebook on the psychological hack it represents:

“When Facebook was getting going, I had these people who would come up to me and they would say, ‘I’m not on social media.’ And I would say, ‘OK. You know, you will be.’ And then they would say, ‘No, no, no. I value my real-life interactions. I value the moment. I value presence. I value intimacy.’ And I would say, … ‘We’ll get you eventually.'”  “I don’t know if I really understood the consequences of what I was saying, because [of] the unintended consequences of a network when it grows to a billion or 2 billion people and … it literally changes your relationship with society, with each other … It probably interferes with productivity in weird ways. God only knows what it’s doing to our children’s brains.”

Talking of mind-bending, last week the newsletter highlighted novel religions obsessed with AI.  This week, Wired profile famed self-driving car engineer Anthony Levandowski who had created “a new religion of artificial intelligence is called Way of the Future” which is intended to serve as a bridge through what he refers to as the Transition to the point of Theosis.  Namely, unification with the AI godhead:

“What is going to be created will effectively be a god. It’s not a god in the sense that it makes lightning or causes hurricanes. But if there is something a billion times smarter than the smartest human, what else are you going to call it?”