We are at this stage of "AI".

Just keep that in mind as you board pressurized Mach 0.9 passengers jets in a few weeks to see your family for the holidays. 42 years after the Wright Flyer, the Bell X-1 broke the sound barrier.

An alternative way to consider what’s happening with LLMs is that they are transforming the *serendipity industry*.
Investors and pundits have been talking about “network effects” forever, but seldom use the word “serendipity”, instead favoring “scale” and “long tail”…
1/N When a platform creates markets that can serve “the long tail” of customer needs, the result can feel serendipitous.

LLMs have mined the *entire digitized corpus* of historical human thought and creativity, and put a compressed version of that on the Supply side of the market.

Given the interest in PyScript and some of the common questions that are coming up, I figured I'd share some of my slides which may answer some questions.

Also I highly recommend checking out our engineering blog post on it: engineering.anaconda.com

The last twenty years of software development have gotten away with not giving a flip about security, really.
Recalling some of the horror stories I've read on HN, reddit, and other forums, it's clear that the entire industry shamelessly plows ahead despite horrible tech debt. This level of blithe carelessness should give all AI & ML practitioners pause.
The lesson of the past is this: Business decision makers will gleefully roll forward with whatever garbage you threw at a wall that happens to stick just long enough for bonuses to be paid out.

@widenka @RealSexyCyborg The long explanation is that Westerners are raised in a cultural tradition built on Enlightenment-era concepts of individualism and liberty as the paramount concerns. This has become even more pronounced in the post-war consumer society of last 50 years. @widenka @RealSexyCyborg Eastern cultures only started interacting with this philosophical mode of individualism about 100 years ago, with the advent of industrialization and labor economics.

So in China, India, and other heavily-socialized countries, ppl can’t just “do what they want”.

The day is here—we’re excited to share findings on the 2020 State of Data Science!

Every year, we check in with the data science community with a survey to see what’s on their minds re: responsibilities, challenges, & processes.

The TL;DR from this year is… (1/5) Hype is cooling down, but there’s still work to be done to help data science achieve business maturity.

Here are my top 3 takeaways:

1️. Turning data into value is difficult. Data scientists report that almost half of their days are spent on data loading/cleansing. (2/5)

My thesis is that since the value prop of ML/AI efforts are complex-valued function ƒ(Software component, Data component), the value of ML/AI companies are extremely non-linear with respect to either partial derivative ∂/∂Software or ∂/∂Data

https://twitter.com/martin_casado/status/1229095693391777792

The disruptive potential of an ML/AI startup depends a LOT on whether the hurdles to predictive success within a problem domain are primarily technological or organizational. Conway's Law applies to data systems even more so to software; and bad data arch can doom DS/ML/AI.

@hawkieowl @glyph Well, I think I understand what you're saying, but would be interested to learn more. Those libs provide intrinsically parallelized data structures, and encourage thinking vectorially. This may be less natural for those who have mostly programmed impressively, (1/N) @hawkieowl @glyph but it's actually a better way to express math/quantitative code. Explicit for-loops are a code smell in most cases when it comes to this stuff.

For the minority of use cases that do fall outside the "vectorized-is-better" space, things like @numba_jit do a great job. (2/N)

@khinsen I think the vast majority of language development is too fixated on the execution of virtual machines, extensibility of the runtime, and expressiveness for higher level modularity. These are 1980s concerns, and they remain unchanged; Java & Python are 30 yrs old now. @khinsen There is an entirely different, under-appreciated branch of potential development centered around data transformation. This basically stopped with Excel. (Anything more complex got sucked into "visual programming" hinterlands.)

@gordonbrander @jamescham Everyone should read "Selling Wine Without Bottles" at least once a year. Nowadays, I reach for it at least once a month. eff.org/pages/selling-…

Go down to the section "A Taxonomy of Information". I'll tease you with excerpts... @gordonbrander @jamescham "information is, by its nature, intangible and hard to define. Like other such deep phenomena as light or matter, it is a natural host to paradox...
understanding of information may emerge in the abstract congruence of its several different properties:"

I emphatically agree with every single one of these.

Python & data science are important to me precisely b/c they are about augmenting human capability across every other domain. They are new literacy for quantitative thought, not just random tech churn.

https://twitter.com/Hillelogram/status/1104909790818566145

Also, VB6 was ahead of its time & it's widely panned/ignored by software devs because it enabled regular people to build useful things for themselves.

It's Shirkey's Law for our industry. We are pro cab drivers and don't see why people would want to drive their own cars.

Utterly naive to call AI a "human right". It's a new battleground. And the Chinese surveillance-state model is going to kick the West's ass unless we get serious about introspecting all the ways we are currently unprepared to even TALK about this problem in the policy sphere.

https://twitter.com/Benioff/status/1088204113228283905

They have better data, on more citizens, and more coordinated action across their entire tech sector b/c it's all government pwned.
Meanwhile in the West, Europe is literally breaking up and the US is *shut down* by an insane man-child.

.@kalefranz succinctly lays out some of the key points in the ongoing conda/wheel discussions that seem to crop up on HN every few months... news.ycombinator.com/item?id=176162…

TLDR: "Vendor the world of binary dependencies" doesn't help anyone. It's pure tech debt. Furthermore: As data science & analytics becomes an increasingly larger portion of Python users, Python itself will be thrown into more multi-language use cases that involve R, JS, Java, Scala, C++... (1/2)

@benjohnbarnes @worrydream @tristanharris @edelwax "scarcity management" is a great term, but markets do most of the work there. Capitalism in pure form is about allocation of surplus, and its model kinda works when economy is tied to *labor scarcity*, and usufruct ties equitably to yields coming from labor and investment. @benjohnbarnes @worrydream @tristanharris @edelwax Tech always puts downward pressure on value of labor. [See Veblen Engineers&Price System] In early days of industrial labor and capitalism, this happened at a rate such that older generations of workers could re-skill or work a lifetime w/o re-skilling.

Not going to speak for @johnrobb but my read is that *networks* circa 2018 differ from mere "connected hierarchies" of the past. As the military is learning in contemplating drone warfare, the only way to fight a network (swarm) is with another network. Unfortunately for those of us who admire/appreciate the utility of existing, classical value systems in maintain societal infrastructure, those value systems are defended by stodgy institutions or stable attractors that developed from a low-bandwidth memetic environment.