Let's be Fwends #114:

Efficiency is Not Your Friend

"When you light a candle, you also cast a shadow." ~Ursula K. Le Guin

We all want to be efficient. But what happens when work never stops, and being efficient means just more work? Today, we dive into the problem with intensifying work, why optimizing for the status quo is a dangerous and risky strategy, and approach a problem with Large Language Models (aka "AI") that is so glaringly obvious we tend to miss it, and have a laugh at people who really struggle with this whole gender-thing. After all that negativity, it's time to re-calibrate our intuition about the state of the world, and see if we can learn a thing or two from very enterprising elephants.

Feeling burned out? Maybe it's part of the plan

In the last couple of weeks, Austria had a very interesting, borderline hysterical discussion about the availability of workers, work time and idea that more people want to reduce their working times (and how austrian employers are ignoring the evidence and just claim that it won't work).

While following this discussion, I learned a great new word. It's Arbeitsverdichtung. Feeling under a lot of pressure at work? Blaim it on Arbeitsverdichtung. Happy when you get a 10 minute pee-break between two online-meetings? Thank Arbeitsverdichtung for that. Struggling to properly close off one task before running to the next? Yepp. Arbeitsverdichtung.

The english term is work intensification, but I somehow like the very bureaucratic and technical sounding german Arbeitsverdichtung more. I somehow imagine a horde of business efficiency consultants and process people walking down an office corridor in lockstep, murmuring "Arbeitszeitverdichtung, Arbeitszeitverdichtung".

What does it mean? Arbeitsverdichtung happens when the same work is done in less time. The "optimal" intensification is reached when the actual intensification is identical with the possible intensificiation. Which is right at the brink of collapse.

So if you feel like you need to do more in less time, and it keeps getting worse, and worse, then maybe, just maybe, it's not about you being a snowflake asking for some work-life balance, but in fact your Arbeit has been verdichtet?

Efficiency is the Enemy

Barry Schwartz reminds us that too much of a good thing can quickly turn into a bad thing, and efficiency is no different.

"At least some inefficiency is like an insurance policy"

Instead of ruthlessly optimizing against the status quo, we should instead headge our strategies against multiple possible outcomes:

"What should we do in the face of this radical uncertainty? When making decisions, instead of asking ourselves which option will give us the best results, we should be asking which option will give us good-enough results under the widest range of future states of the world."

Setting up a system that can work in a multitude of environments seems to be superior to creating a maximally sufficient system and hope that you can pivot it fast enough.

Everything AI knows is from the Internet. And that's hardly a good thing

Melissa Heikkilä tried out a popular AI app for creating online avatars out of selfies just to find out that it picked up on her Asian heritage and created tons of sexualized images of her, probably based on Anime and similar depictions of asian women.

"Because the internet is overflowing with images of naked or barely dressed women, and pictures reflecting sexist, racist stereotypes, the data set (of AI models) is also skewed toward these kinds of images."

The problem with those "Large Language Models" (as they are called) is that they are trained on scrapes from the Internet. And while we might assume that they have many safety mechanisms installed for various obviously dangerous content, they will still take everything at face value. They have no outside world, no other reference frame. For them, the Internet is the world.

You can imagine playing all kinds of tricks to a mega-brain confined to such a specific, biased realm of knowledge and information.

Or, as Gwen Branwen puts it:

"A reminder: a language model is a Turing-complete weird machine running programs written in natural language; when you do retrieval, you are not 'plugging updated facts into your AI', you are actually downloading random new unsigned blobs of code from the Internet (many written by adversaries) and casually executing them on your LM with full privileges. This does not end well."

Source

That Gender-Thing is hard

Yeah, I know, this whole gender thing is hard. So who can blame the people responsible for those online forms for not getting everything 100% right?

Personally, I like "Boing 787-9 Dreamliner".

Seriously, many things are getting better

Sometimes it feels like things go down the drain. And some certainly are! But not everything is taking a turn for the worse, there are actually some areas of development that are actually improving. How does your intuition compare to the facts? Use The Shape of Change to find out.

Elephants who know how to game the system

In an act of elephantine banditry, these big animals use their right-of-way to halt trucks loaded with sugarcane, have a snack, and then let the vehicles continue.

I just love it when someone takes a rule that is already in their favor, and goes "... how far can I take this before someone stops me?"

That's it for this edition of Let's be Fwends, may your Arbeit be barely verdichtet, your sugar canes safe and your days pleasantly and refreshingly inefficient.