Inkwell: Authors and Artists

    
The World Economic Forum has a study out that predicts at least
"elevated risk" of "global catastrophe" within the next 10 years.

How is that for state of the world?

https://www.weforum.org/publications/global-risks-report-2024/digest/
  

    
<doctorow>,
> understands radioactive decay

My understanding of radioactive decay and of chemistry is that
uranium and many of its decay products are poisonous, and that they
leach out of glass.
  

    
Post <222> Via email from Shebar Windstone:

> I'm socially challenged when it comes to facial expressions & body
language. Not to mention remembering names & faces. Could AI help me
there???

I am a prosopagnosiac (face blind). Google glasses with facial
recognition would be a huge benefit. Facial expressions are a bit
harder, but yes I'm sure it could. Body language is even more
difficult and cultural, like the Indian head-bobble.
  

    
Emotion recognition: can AI detect human feelings from a face?
<https://www.ft.com/content/c0b03d1d-f72f-48a8-b342-b4a926109452>
[ registration required - failed for me ]

I recently read a "law" about headlines : if there's a question mark
then the answer is NO.
  

    
>"Any headline that ends in a question mark can be answered by the
word no."

Betteridge's Law, to wit: "Any headline that ends in a question mark
can be answered by the word no." Coined by Ian Betteridge, AKA
<ianb>.
  

    
We sometimes think we live in a world that's solid and permanent,
but that's not the case. The state of the world is fluid, flowing,
impermanent, often hard to grasp. Our experience is local but its
reflections are global, universal. 

As I write this, I hear an owl outside, hooting into the wind,
asserting. It's the time of year for the owl to nest nearby and
claim territory. We're here, sharing that territory, what can the
owl do? 

Northwest winds 25 to 35 mph with gusts up to 55 mph expected
tonight - will the owl blow out of the nest? It's a local problem.
The world won't notice.

Over the weekend, temperature here will drop - by early next week,
temps will fall into the teens, 40+ degrees below the usual for this
time of year. Weather is unstable - but weather was never permanent.
It flows, it changes. Climate change creates greater instability.
The world may be unlivable a few years from now... but I can't know,
I can only wonder. 

Months ago I hopefully joined Rushkoff's Team Human, encouraging a
commitment to humanity over technology. It's fine to have AI, but AI
has its place. It's fine to have technology, terrible to use it to
abuse and exploit. 

"Why Practice Loving-Kindness?

"Through metta, we become more honest and loving toward ourselves
and the world around us. In metta meditation, we wish happiness,
safety, and ease toward ourselves and others. In the most common
version of metta practice, we offer these feelings of goodwill
through the use of simple phrases first to ourselves, and in
succession to someone we love, someone we're indifferent towards,
someone we consider a "difficult person," and finally to all beings,
everywhere. This allows us to begin our loving-kindness practice
with feelings most accessible to us, and work up to more difficult
feelings, such as sending love to someone you have a negative
relationship with."

https://www.lionsroar.com/metta-meditation-guide/
  

    
Does the owl hooting at your gate predict the future of the state,
of the world? Who knows.  

If I may, I'd like to pull on a thread in <231>:  mindfulness, and
its related categories.  It brings up one of three topics I'd flag
before we all pack up our tents, saddle up our camels, and move on. 

Ron Purser is partly right.  Mindfulness has become the new
capitalist spirituality. However you parse that, mindfulness has
been a key element in many Buddhist schools, thus nothing new, in
and of itself.  But its adapting to, and adoption by, Western
psychology (& neuroscience) is news; first in the health field, next
in education, now at the workplace.  

It was never a topic (to the best of my knowledge) in WER, but
deserves to be considered a topic to watch in SOTWs ~ along with
religion(s), in general, which change slower than states, who move
slower than culture. A broader p.o.v. is the context of more and
more people making spirituality real for themselves, in everyday
life, up out of the Sunday pews, or Saturday synagogues, for the
past 25 years now.

Another strange attractor here is the continuing pull of the
fundies: besides the upcoming election, consider their powerful
influence upon Israel (see the film 'Til Kingdom Come [2020] for
instance).

Islam?  Bruce and others have tales to tell from their experiences
in Yugoslavia, I'm sure.  This moment in time, I'm thinking of the
lightning rod which Israel's response in Gaza is forming for Muslim
unity.  How will that manifest, down the line ... ? 

   
  

    
My other topics I'm watching, year by year, are war & economics, not
necessarily separately. Economics is the continuation of war by
other means, right?

Just as the number of nations playing The Great Game have expanded
from Britain and Russia and central Asia, so too are the BRICs
growing. This year, new members to the club included 
Saudia Arabia, Egypt, the United Arab Emirates, Iran, and Ethiopia. 
(Does anybody smell gas?)  Doesn't it spell further threats to the
perpetuity of the petrodollar (circa 1973)? Another economic
development from Brics, ten years ago, was the New Development Bank,
proposed by India at an annual meeting. The list goes on. Since we
don't know what's backstage, let's see what's on their table when
they meet in October.

I'm, again, a frog at the bottom of a well, who only sees a few
stars, hardly the whole night sky.  All of the above is open for
SOTW questions, comments, and criticism, ad libitem. 

 
  

    
It seems rather odd to me that, when things were bad, we were more
curious.

Take the pandemic. This was good reason to be curious about all
sorts of things: epidemiology, genetics, FDA regulations, Chinese
biosafety practices, Russian vaccine manufacturers, bats, minks,
vaccine supply chains, toilet paper supply chains, electronics
supply chains, container port logistics.

Or how about the Ukraine war? I remember being pretty curious about
internal Russian politics, learning a fair bit of Ukrainian
geography, reading military expert assessments and detailed
investigations of gas pipeline explosions.

You don't have to *like* things to be curious about them. You don't
need to be in a good mood.

What is there to geek out about now? What are you curious about? How
am I going to get my news junkie fix?

I've ended up reading a long history of Ivory Coast, of all things:

<https://mattlakeman.org/2024/01/01/notes-on-the-ivory-coast/>
  

    
Well, there's always Matt Levine as a source of stories about
financial shenanigans. Apparently the Fed is loaning out money at
4.93% and paying 5.4% in interest. Banks have noticed and are
getting the free money while they can.

Markets Get Ready for Risk-Free Fed Arbitrage Trade to Expire
<https://finance.yahoo.com/news/markets-ready-risk-free-fed-205248821.html>

> The Bank Term Funding Program, which supported lenders through
that tumult, “really was established as an emergency program,” said
Michael Barr, the Fed’s vice chair for supervision, at an event
Tuesday. The BTFP, which is slated to expire March 11, has attracted
a swath of users in recent weeks as traders look to lock in loans at
a rate that’s more attractive than the alternatives.

...

> Usage of the program — which allows banks and credit unions to
borrow funds for up to one year, pledging US Treasuries and agency
debt as collateral — reached a fresh high of $141 billion in the
week through Jan. 3, according to the latest Fed data.

> That’s because the BTFP rate, calculated as the one-year overnight
index swap rate plus 10 basis points, has come down as traders
boosted bets on more rate cuts in 2024. Eligible institutions have
found it cheaper to borrow cash through the nascent facility, paying
about 4.93%, with the opportunity to park that cash in an account at
the central bank that earns 5.40%.
  

    
I'm glad that my cocktail-glass choices are meeting with such public
approval.  After spending time in Italy, I've come to understand
that making a whole lot of ritual fuss about alcohol consumption
makes you drink *less.*

You do waste more time and money on it, and you discuss it lots more
-- "Why do you drink that Barbaresco, when for a mere five times the
cost, you could drink good Barolo?!" -- but your liver doesn't
corrode, which is a distinct plus.
  

    
During my extensive stays in Turin, I've seen a fossil-fueled
motor-vehicle town slowly transforming into a tourism center.   In
practice, that meant FIAT (somehow renamed "Stellantis" nowadays,
although that proved about as popular as calling Facebook "Meta")
slowly handing over its stern cultural dominance to the "Slow Food
Movement" and its retail arm, The Eataly grocery chain.

Food tourism in the Turinese region is known to its adepts as "Il
Food," because it's sold to visitors who don't speak Italian and
don't need to know what Italian food actually is.   So  "Il Food" is
that hundred-year-old "Americano" cocktail I mentioned earlier, but
it's industrialized and rationalized, and aimed at over-fancy
value-add edible products, with conspicuous profit-margins, which
are spread from soup-to-nuts, with toney, ritualized four-star
restaurants taking the place of cheap cafes. 
  

    

 "Il Food" is made by Italians with Italian ingredients, but it's
not what Italians themselves would choose to eat.   It's quite 
similar to fancy, erotic Italian lingerie, which Italian women don't
normally wear;  anxious housewives from Bremen are supposed to buy
all that stuff.  

A good example is Turin's time-honored "Lavazza" coffee  breaking
out with their own town museum in Torino, and doing peculiar,
specialized, one-off brand-extension coffees, where the packaging
has long explanations in English.  They don't drink that themselves;
they promote it.
  

    
So I watched "Il Food" more or less taking-over Turin; the foodies
were the gray-eminences and sponsors of pretty much everything, and
if you ever had an event with foreigners, they were sure to show up
with the catering.

Last year, though, I really saw the art-scene just blow up in Turin.
They've always had one -- "Artissima," "Paratissima," lots of museum
and gallery traffic -- but last year the scale of it was colossal. 
Huge street crowds to gawk at the art, carnivalesque, even.   It
felt like their art-production and art-display racket had been
motorized.   Even the Turinese  veteran art mavens were startled and
were asking each other if they'd  somehow turned into Florence.
  

    

In Turin I'm a guy who spends a lot of time within the art-world
demographic, so of course I'm pleased by this development.  I would
rank this as a positive developmnent fror 2024, something pleasantly
surprising that I look forward to.  I don't yet know how a
tourist-centric "Il Art" is supposed to pay off for the Turinese,
but a flauntingly artistic, aesthetically avant-garde Torino, I have
the warmest feelings about that. 

 So much so, that I'm thinking that the time comes when you don't
need to push that avalanche yourself, and it's more use to step out
of its way and watch it roll downhill from the Alps.
  

    
Here in Austin we're preparing as well as we can for another "polar
vortex" event. Bitter cold in the east is less of a problem, because
those folks know how to deal with it. But down here in the Texas
tropics, we don't normally or readily do cold. It's just not part of
our experience, most of the year.

The polar vortex of 2021 almost shut Texas down. The energy
infrastructure couldn't handle the cold, and because wily Texans
avoid federal involvement be operating an isolated one-state power
grid, we couldn't get help from outside the state. I suppose that's
consistent with the fantasy of rugged individualism in this state -
a bunch of cowboys who are independent, strong, stupid, and frozen.
But those cowboys don't exist, and the citizens of Texas were plenty
pissed when the grid collapsed and the water systems failed. We
filled our bathtubs with snow for water to flush the toilets. Other
people had it much worse - some people died.

The powers that be tell us that they've fixed the infrastructure and
can handle the cold, but at least rolling blackouts wouldn't
surprise me one bit.
  

    
I'm curious about how you are preparing this time?
  

    
Mostly stocked up on water and canned food, but honestly, it's hard
to prepare well enough for loss of both electricity and water. I
like to believe that was a freak occurrence that won't be repeated.
I did consider buying a generator, but never went through with it.
  

    
Haven't bought one, but there are a lot of big but portable lithium
batteries intended for camping these days.
  

    
If anyone is curious about how Russian military production is doing
on ships, planes, and tanks, here's a good summary:

How did Russian defence industry perform in 2023?
<https://wavellroom.com/2024/01/10/how-did-russian-defence-industry-perform-in-
2023/>

To sum up the summary, production isn't enough to make up for
losses.
  

    


John Ohno writes:



Regarding the sentiments expressed in <241> and elsewhere:

One effect of time-binding is that, the more you know about history, the less
surprising "the future" ends up being, and our society is one that now time-
binds everything. We used to be able to kid ourselves that the telephone was
invented by Alexander Graham Bell, but now it's impossible to ignore that
Bell simply happened to be one of several guys doing incremental work in a
particular field at exactly the time that other developments in other fields
made certain kinds of breakthroughs possible. Everything of substantial scale
that happened this year has deep roots in previous years that made it
inevitable or nigh-inevitable, and next year will be the same; it's only
that, for the most part, everything substantial this year was something we
*already knew about* last year.

A lot of the discussion in this SotW was about the two technologies that come
closest to seeming "new" (at least in terms of things we mere mortals are
allowed to play with): GANs and LLMs, which we have for some reason begun
collectively calling "AI" even in circles that ought to know better. Both
existed in previous years, and maybe that's why we can't bring ourselves to
be too excited. Both are built on a set of mid-teens technologies (tensor
flow and friends) to run a 90s technology (neural nets, backprop optional) on
a naughties technology (GP-GPU).

The GAN is a genuinely impressive improvement over its immediate predecessor
("deepdream"), and blows other mechanisms for constructing arbitrary images
from plaintext descriptions out of the water; because of GANs, regular people
are considering the corner cases of copyright and trademark law (though they
are not, as a rule, actually informed about this subject). At the same time
as some illustrators are loudly complaining that they should be getting a cut
of profits from all GAN-produced images (due to the size of the dataset, I'd
expect most people's cuts to be on the order of a few cents across a
lifetime), other illustrators are quietly integrating GANs into their process
(often replacing photobashing -- which itself collages a bunch of images not
owned by the illustrator's employer to form the basis of a new work).

LLMs, on the other hand, are part of a very solid lineage. When you go about
trying to figure out how to make a computer emit text for human consumption,
there are several paths: you can post-process existing human-generated text,
or you can construct new text based on a model of what text ought to look
like; models are either constructed from the ground up or are statistical. We
used to call this the "neats" vs "fuzzy" dicohotomy in AI research. LLMs are
"fuzzy" and the technology here is pretty typical of statistical text
generators (some of which can be very simple: a first order markov chain, for
instance, does a great job of mimicing a very stupid, possibly intoxicated
human). This is old tech -- markov models date back to the 1920s and pretty
much as soon as we had computers we started using them in conjunction with
markov models to produce poetry. "Neat" text generation isn't much younger:
computers were producing crappy poetry from templates in the early 50s,
writing crappy screenplays in 1955, writing fairy tales based on complex
models of worlds filled with actors with motivations in the 70s (TALESPIN),
and writing published novels in the 80s (RACTER, which was a hybrid of
templates and markov models). If you want to understand what other
technologies exist and how they stack up to LLMs on various metrics (such as
"can I run it on my old laptop", "would I actually read 90k words of this
crap", and "does it represent a kind of performance art critique of modern
society"), take a look at the archives of NaNoGenMo, a project where for the
past 10 years people have been using various techniques to computer-generate
novels every November.

What else do these technologies have in common? They're both statistical, but
also, they both depend upon very large corpora over "smarts" in the actual
code. (This dismissal of "smarts" is actually probably a serious problem:
there's a paper going around indicating that using the A* pathfinding
algorithm, something taught to all CS undergrads, makes GAN image generation
much faster on CPU than it is on GPU, yet nobody is picking the low-hanging
fruits in practice because cycles are treated as cheap -- yet cycles are not
*actually* nearly as cheap as this, and we are beginning to pay interest on
all the deep-discount cycles we thought were deals but were actually loans.)
The large corpus is also a major problem. Large corpora are tough to do
*anything* with (personal experience: the US patent database has, last I
checked, thousands of unique misspellings for the patent assignee "IBM"), and
owners of the models claim that checking copyright is just too hard. This is
probably legitimate: copyright registration is optional in most countries, so
nobody actually has a list of all copyrighted materials, and registered
copyrights may themselves be illegitimate, so the only way to truly verify
that something is copyrighted it to challenge its copyright status in courts.
Large companies have been accumulating their own portfolios of copyrighted
work for ages, though, and some of the biggest may have enough to "own" their
whole models. More difficult though is relevance: the sheer volume of the
minimum corpus size necessary to learn useful patterns in an LLM or GAN means
that the corpus will necessarily also contain useless or actively harmful
patterns. OpenAI and others have been trying to make those patterns
inaccessible by automatically adding constraints to user-supplied prompts,
but these constraints are subject to the rules of human conversation as the
neural net models them, which means you can wheedle and bully any statistical
model into producing whatever output you want.

People argue about whether or not the output of these things is sufficiently
"new", seeing as how the training corpus is all already-existing stuff. I'm
more interested in whether it's "novel" -- and the bigger the corpus, the
more the individual idiosyncrasies of individual contributors to that corpus
get evened out. This is great in one respect: professional illustrators and
writers of ad copy don't actually want to be too arty, so output that is so
generic that nobody could possibly try to trace copyright provenance is just
what the doctor ordered. But on the other hand, the genericness of a large
training corpus infects everything, to the point where when somebody got a
bunch of LLMs to simulate important historical figures in order to teach
history to students, the one mimicing Hitler was unwilling to be anti-
semitic. (Hitler's antisemitism as it collided with his political power is
probably the most important thing for a student to know about him!) We could
make an argument that LLM and GAN output is holding us back by presenting a
mosaic of fragments of the past as though it is the present -- and some
people do -- but I think time binding does that by itself; in my view, the
problem is that the corpora here are too large and the output too good to be
anything but boring. Talking to an LLM is not like talking to an alien form
of intelligence, but rather, like talking to an extremely stupid employee
whose eagerness well-outstrips his skills.

It is not surprising that, for instance, technical managers are much more
interested in the possibilities of LLMs for software development than
software engineers are: getting an LLM to produce working code is very
similar to getting an unpaid intern to do it. But, LLMs are completely ill-
suited to maintenance tasks (the majority of software development work) so
they probably will not take off in this domain.

People are using LLMs and GANs to produce reams of crap, which they then try
to sell, but this is stupid: there are cheaper and easier ways to produce
reams of crap. Even the scammers are being blinded by hype.

So, we've established the historical tail of modern "AI". Other current
events from this year: the inevitable war in Israel, which we've all been
waiting for since the 70s, and which looks like it'll unfortunately go the
way we've all expected (a minority population exterminated by a modern state
who has convinced itself total war is justified); we spent the year mid-
rampup to another american election with no third-party upset or meaningful
party-policy change in sight; major social media platforms continue get worse
to the degree that the term used to describe this becomes "word of the year".
Bad, boring, all with strong ties to previous years, and no particular sign
of changing in 2024.

The social media thing is particularly interesting to be, because it also
looks very much like a problem of ignoring for failing to evaluate low-
hanging fruits. Many people left Twitter, but most of them ended up on
Threads (some kind of mashup of Instagram and Mastodon) or Bluesky (Twitter
with Mastodon's federation-derived username structure but without Mastodon's
federation) instead of Mastodon (which I've been on since 2017, and is part
of a network that dates back to when Twitter was new). People claim that
"mastodon is too complicated", but its interface looks and behaves very
similarly to twitter, and every social-media task that is important to
individual social-media users (as opposed to tasks that are heavily used only
by professional influencers and their ilk) operates more or less identically
with twitter. Meanwhile, the Mastodon people ignore and look askance at
Secure Scuttlebutt, the only social network I know of that takes
decentralization seriously (it's fully peer-to-peer). Unfortunately, the
Scuttlebutt people also aren't taking full advantage of it: there are several
implementations, but most are unmaintained, and the fully-functional ones are
electron apps for some unclear historical reason, making them too bloated and
unreliable to keep open (a difficulty for a social network that only syncs
with currently-online peers). We're surrounded by the possibility of serious
technical innovation -- the kind that produces cultural innovation -- but
instead, we are too tired and burnt out to even try to understand the new
technical innovation, and we just hold onto the old one as it gets worse and
worse; meanwhile, the people working on the technical innovation don't care
that they've written in a way that limits its utility and user base.

For all of 2023, carrying into 2024, I've been worried about these trends.
I'm powerless to do anything about them on the macroscale, but I am trying to
combat them on the microscale: can I make sure to read books that are not in
anybody's training corpus, that nobody I know has read; can I try to get some
people to look at these promising paths, and maybe travel some of them
myself; can I try to make a window to escape through? The enemies of the
future are powerful, in part because they like to believe they are friends of
the future. They are friends with a future in which I no longer exist, and
they probably wouldn't either.
  

    
I had to search what GAN is : Generative Adversarial Network

Bard: GANs pit two neural networks against each other: a generator
crafting novel data (images, music, text) and a discriminator trying
to catch its fakes. This adversarial training loop pushes both
networks to excel, generating increasingly realistic and creative
outputs. They learn not just from data, but from actively
challenging each other's skills.

[ Iterative session with Bard, refining what I wanted for this post.
One (too long) was very creative, so I had to ask for shorter, more
technical ]

But agreeing that in many ways 2023 was a break-out year of things
in the pipeline.

Anyone interested in a new pangolin coronavirus that is highly
infectious and kills 100% of infected human-ACE mice? In 8 days? By
destroying their brains? Now if THAT escapes from its chinese lab
....  
  

    
> People claim that "mastodon is too complicated"

Which translates as "I couldn't figure out which server to pick."
  

    
Took three tries for me.
  

    
On AI/LLM's I'll wrap up my <ai.ind.> comments as :

Caveat Promptor