omenti

Das ist nicht nur nicht richtig; es ist nicht einmal falsch!

The Office of the United States Trade Representative released the formula used to calculate Trump's "reciprocal tariffs.

My gosh!  Where does one begin?

The famous physicist Wolfgang Pauli was asked to look at a paper by a young physicist.  After a few minutes he tossed it away saying: "Das ist nicht nur nicht richtig; es ist nicht einmal falsch!"  (That is not only not right; it is not even wrong.)

 

There are constructions so completely without clue as to merit Pauli's comment:  the flat Earth, the Earth is the center of the universe, phrenology, or that π can be legislated to be 3.2  come to mind.

 

Trump's formula completely ignores services, foreign demand and a host of other important items. His real goal is probably more along the lines of a protection racket - a tool to compel fealty.  We shall see. I can imagine people around the world acting like Canadians have when it comes to American products and services.  Not to mention the psychology of feeling "poorer" in the US. 

 

What could possibly go wrong?

_________

Catching up.. five readers have written enthusiastically about the maapl.net search engine. No AI, no ads, open source, no tracking and free.   It's also faster than Bing and Google.  I've been recommending Kagi - a paid search service - but probably won't renew my annual subscription when it comes due this Summer as maapl fills most of my needs.

 

Finally congratulations to Kip, Andrew, Frank and Jheri who solved the puzzle I posted on the 23rd of March.

 

 

 

 

 

 

 

getting lost for the thrill of it

 Last week a friend and I were chatting about the tension between exploration and exploitation in technology.  The work of James March came to mind.  My old friend Norm had a few of us read his paper on exploration and exploitation for a few of our Friday afternoon sessions at Bell Labs.  It had set off a good deal of work in a number of fields and was relevant to the Labs as exploratory research was cut in favor of more exploitative work. March's paper was interesting as it suggested ways to model the tension between research, development and exploitation when resources are limited (they almost always are). It also brought up the fact that people have differing tolerances and tastes when it comes to the risks associated with each of these areas.

 

For the moment I'm more interested in concentrating on exploration as that's what I find most rewarding.  By many metrics individual exploration isn't worth it compared to sticking to a well-trodden (by you or your firm) path.  It's difficult and sometimes impossible to optimize and it frequently fails.  On the other hand if you're willing to learn from failure it can be powerful and an education plus every now and again you have an insight or discovery. 

 

Some of  us find the discoveries or insights - and even the path of being somewhat lost - fun and even thrilling.  It's been suggested there's a dopamine hit that helps drive this in some.  It's not something we can do all the time as it's generally not productive, but the discoveries and flashes of insight are what can bring change.

 

There's a good deal of work that suggests interdisciplinary at the individual and/or organizational level provides richer explorations with higher success rates.  It certainly seems to be the case in the physical sciences.  Those who make the greatest strides are often those who have come up through varied pathways.  Creativity appears to be linked to interdisciplinary as well as focus.

 

Doing something is much richer than just watching.  It's been noted dementia rates are much lower among those who are constantly doing things that require learning.  It doesn't seem to matter what it is, but has to be something that's mentally challenging.  Learning to knit, a sport you haven't done, math you don't know how to do, how to draw, play a musical instrument .. anything as long as it's hard.  

 

There are false novelties.  Social media gives a stream of novel things to watch giving people dopamine hits. There's a bit of an addiction. Perhaps we're wired for it. Another false novelty is sugar.  When we needed the calories sugar provided a hit.  Now we don't need the extra calories, but for many of us the attraction of sugar can be unhealthy.

 

Back to creativity.  I might be wrong, but I've come to believe it's something you develop and enhance.   There are several measures of it.  You have to take them with a grain of salt as it's not exactly clear what's being measured, but there's a general agreement among these methods that creativity has been dropping in teenagers and college students since the late 90s with a very large drop beginning around 2010.  A sobering conjecture is heavy use of social media is taking away from other forms of the work of exploration.   It's also been noted that this drop isn't seen among those who are doing tasks that require constant exploratory learning.

 

So turn off the GPS every now and again and take find new paths.  Surround yourself with interesting people who are dissimilar.  It's something of a luxury to do much of this in most work environments so take up new hobbies and keep pushing yourself.  

 

 

creating the arrow tips

To start off you need to subscribe (free) to the Crazy Stupid Tech newsletter from Om Malik and Fred Vogelstein.  A couple of paragraphs from their introduction gives a sense of where they're going:

 

Together we have followed Silicon Valley’s innovation engine for decades. We've seen a lot. But one observation stands out: The best ideas - the ones that launch meaningful companies - need to seem crazy and stupid at first.

We get that tech makes many of us more nervous than hopeful right now. But as Silicon Valley’s influence extends deeper into all corners of our economy and lives, there also have never been more interesting, crazy, stupid new and interesting technologies out there. They are not getting enough attention. So our goal is to tell you about them.

 

I've been involved in a variety of technologies from application to tools, but I'm a bit of an outlier.  My background is fundamental physics - something I've been able to mostly keep up since grad school while doing other things and what consumes most of my time these days.  It's non-linear and the payoffs often don't arrive for decades and rarely for those who did the work.  But over time it's where almost all technologies come from as well and a number of powerful tools.  It's difficult to quantify the payoffs, but most numbers exceed one hundred time return in investment after factoring in inflation - and that includes a large number of paths that didn't work out.  It may come as a surprise, but this is a case where the liberal arts win over time.¹

 

A secondary effect is the training.  Get to the point of a postdoc in physics and you begin to think like a physicist.  The same for chemistry and probably many other areas.  Sometimes this ability to 'think different' is useful in more applied areas - particularly when it comes to stepping outside just a bit.  Of course you probably don't want a room full of physicists, chemists and mathematicians working on finance .. well .. unless you happen to be Jim Simons, but he was probably an outlier. 

 

This kind of research, often called curiosity-driven or discovery research, has been increasingly under fire in the US and EU for about thirty years. Funders are much more interested in some kind of payoff - more of an applied research (although that term is used for something else in industry).  This has happened before and I suspect we'll see a swing the other way as China takes over as the leading scientific power.

 

I'm not discounting applied research.. more of it needs to be done and I suspect a better focus is required.  The enormous sums of money and numbers of people going into computer science for gAI strikes me as mis-targeted and wasteful given more pressing challenges around global warming, inequality in healthcare and so on.

Finding the right mix of people and research areas at the fundamental end demands a great deal of good taste.  A few organizations have been very successful: Cambridge was behind in sorting out issues of telegraphy and the likelihood of radio, but decided to work on fundamental issues of thermodynamics.  They managed to assemble some of the best minds and created the modern research laboratory.  They created the spark that launched quantum mechanics which exploded in the 1920s.  In the teens Caltech was a sleepy little engineering school that found itself with a new mandate.  They decided not to work on  the transmission of electricity like everyone else.  By going deeper they managed to become one of the more relevant pure research organizations for the next century.  There are other examples like Bell Labs which had about six decades.  To be fair it was mostly applied research, but there was a core that changed the world.

 

There's an interesting form of teamwork at these places. People tend to talk to each other a lot.  Places like the Perimeter Institute have blackboards in nearly every room and hallway.  Each table in the cafeteria has notepads and pencils to facilitate the discussions that flow across groups and through the building. Setting direction is an art that is often based on a perception of beauty. It's similar, but slower in the experimental laboratories and there's something of a competition and balancing act between theory and experiment.   It may seem like herding cats and it's messier and more inefficient than the so-called scientific process, but it turns out to be the most efficient way to create possible futures.  It's an act of faith that learning more about the Universe is important to humanity.  Of course it takes time along with the insight of people with different skills to turn that into reality and Crazy Stupid Tech will probably give some insight into that direction.

__________

1 The physical science and math are part of the liberal arts as opposed to professional and technical fields. 

 

 

finding your tools

In the tenth grade I asked for a single Christmas present.  At nearly thirty dollars it would be larger than normal plus there would have to be an upscaling of my sister's present.  I wasn't hopeful, but it actually happened. A Pickett N4-ES slide rule.  I spent an enormous amount of time playing with it.  A few things come with heavy slide rule use:  the ability to do estimates in your head - particularly involving orders of magnitude and an instinctive feeling for how logarithms work. Both of those skills are still with me although I rarely use a slide rule these days.¹  There are other tools that you develop a feeling for through a lot of play - Mathematica and Maple come to mind.

 

 

We all use tools.  They're often connected with and change with technologies although some of the simple good ones never seem to change much.  It's interesting to sit down and make a list rating them on a love/hate scale.  

 

Some of us build our own.  Engineers excel - a good friend was working on a certain class of problem in computer science when he started at Bell Labs.  His director took him aside and suggested maybe he should create a computer language to give himself the tools to work on the other problem.  It turned out building and improving the language became his primary focus and a much richer path than the original problem area.m That language is a tool used by millions of programmers every day.

 

And there are tools that have caused frustrations to millions.  Too many user interfaces are great examples of not understanding fundamental issues. Automotive touch panels are a favorite example of mine. They don't provide feedback and require too much attention from the driver.  I was delighted to learn cars with these interfaces won't get good safety ratings in the EU beginning next year.  Perhaps we'll see a some thoughtful physical interfaces.

 

In physics Feynman diagrams are important. In the late 40s and early 50s he and a few others revolutionized the field with quantum electrodynamics which led to more general field theories.  The most accurate theories in all of science describing much of the universe.  Feynman was disturbed by a few issues and wanted something that felt more Newtonian.   He came up with a way of describing interactions that could be summarized with what looked like doodles. The approach wasn't the right approach, but the diagrams provided a way to make calculations much simpler and to think about underlying processes.  Simple calculations could be done thousands of times faster and more accurately and, more importantly, the technique shifted how people think.  Frank Wilczek wrote an excellent description in Quanta. 

 

Of course computational hardware and software of all types have become important tools individually as well as embedded in complex systems.  We tend to find those tools that suit our own needs.  What you find useful and even delightful may not be on my radar and visa versa.   I tend not to recommend much to people when I don't know their use cases - which leads me to the reason for this post.

 

Several people have asked for primers as well as my take on generative AI (which many simply call AI, although that's inaccurate).  I'll skip the primer - you can find those in a hundred other places, but I've had a bit of experience.  In the past year I've served on two committees that made recommendations on gAI student use in college.  I think we made some good progress, although it's clear there will be changes over time.  I've also used several of the tools - free and subscription - and have formed some initial impressions.  I'll break them down into useful, take it or leave it, and not worth the effort.  

Useful

Non-technical translation. Corresponding with people who don't speak English and reading posts and articles.  There are mistakes, but there's enough context to get the gist of what's going on

 Scraping text and even equations from images.  (the text part is built into OS X and iOS)

Transcription .. some errors, but again context saves you

Entertainment - I've seen people lacking artistic skills have a lot of fun making images.  The same for non-writers looking for frameworks to start.

Some of Apple's local gAI looks useful with the emphasis on local, but that's still a work in progress and I have no experience.

 

So-so

Formatting LaTeX .. it's far from perfect, but save a bit of time. 

Translating literature .. something is lost over a good translator.  There's also a good deal of sexism and racism 

Playing with ideas as a part of learning.  I haven't used this much, but ran into it several times in gAI student use. I'm on the fence, but it might move up to the next level.

 

 

Not so good

Search - what were they smoking?

Technical programming..  it takes more work to debug and integrate with existing software than writing from scratch. 

Summarizing anything technical..  it's often difficult to find the mistakes as it "lies" smoothly

Physics and math problems.  It gets well-known examples, but beyond that is wrong

Re-writing pieces..  some love this, I found it a big fail. 

Some bad experiences with gAI chatbots

A number of ethical and energy use/pollution/privacy issues.  I think there are large risks associated with LLMS . not super intelligence in the guru, but serious risks now.

 

I'm no longer paying for a subscription or actively using services like ChatGPT.  I'll be following along, but I wasn't able to find a tool worth my time on a regular basis .. at least not enough to counter the ethical issues I see.   Of course I recognize everyone will have their own cases.  And back to the slide rule - to find if the tool really fits, you have to spend a lot of time playing with it!

 

I find a few other flavors that come under the heading of AI extremely useful!  I'm very curious how many people will pay serious money for gAI tools?

 

 

__________

¹ I'm of the age where the best things I've owned need to go to close friends and old students.  The slide rule went to a student of mine under the condition that she learn how to use it.  She did and still does. 

(way) beyond black swans

a mnipost

 

Fundamental change comes from discovery - something that science is wired to do.  The process of science is admittedly messy and a bit inefficient, but it works and has vastly expanded our understanding in so many areas. There's this myth about how it works -  the scientific method kids had to memorize in school is a not very accurate model of how things really operated - at least the areas I know. All of the pieces are there, but they're generally non-linear and can spread out over long periods of time.   What drives people are the rare magical moments.  For most of us they're rather small, but the pleasure in discovery is enormous.  It's very hard work as it's so easy to fool yourself.  Some of the most dramatic discoveries come when you're looking for something else - if you're careful enough to recognize them. 

 

One of my favorite examples is a 24 year old astronomy grad student who fundamentally put a dent in astronomy and physics.  Her discovery involved a huge amount of careful work.  Her advisor was awarded the Nobel Prize in Physics while she got nothing -after all - she was a woman and a student in the 60s.  I've been lucky enough to have had dinner with her - she sees her lack of the Nobel underscoring how sexist science and the Nobel committee are.  

 

The interview is excellent and something to show to kids for inspiration as well as dealing with the imposter syndrome.  She's done so much for women in science.

 

creativity in film and beyond

Unburdened by the plastic wrap so common to art books, it was sitting  in the bookstore and was reminded how much I wanted it.  A copy of The Art of WolfWalkers. Ten minutes of delight, but it was too spendy for this trip.  

 

 WolfWalkers is the third in a set devoted to Irish myth from Cartoon Salon - a tiny studio in Kilkenny, Ireland that has won respect from Pixar, Disney Animation and Studio Ghibli.  It's on Apple TV+ if you like amazing animation. 

 

Animation studios and art schools are amazing places.  Years ago they seemed like black boxes that took stories and turned them into animation.  More recently I've been involved with one and am wobbled by the amount of creativity and work play that goes into the most simple moments.  It's always a team effort spread among disciplines and years.  

 

After a few months I finally bought the book. Just wow!   Here's a video of a roundtable the before the film was released.  It was my introduction to the film (although not to the myth or its historical offshoot as Kilkenny is an ancestral home). 

 

It's worth writing about as (a) some of you may enjoy and even love it and, (b) creativity in the era of generative AI  is something I've been spending time on this past year an animation studio and two universities.  Signals have emerged. 

 

The New Yorker published a must-read piece on AI and Art by Ted Chiang.  Given my experience with animation studios, a music conservatory and physics and math, I'm in conditional agreement.  I do see a place as a tool, even in the creative arts, but more as a tool to relieve drudgery rather than become 'thought'.  I use it to organize references in papers as well as to create simple programs that involve freshman level programming for simple but time-consuming tasks.  I stay far away when precision and accuracy are needed and I don't see the creative spark. 

 

Film studios, musicians, and artists worry about wholesale theft of their work in the short term and the potential, if these tools improve, that we'll see more 'art' of lower caliber largely because human-created art is too expensive.  I think the later isn't possible with generative AI - at least not for quality work.  That doesn't mean real art won't be forced out by economic pressures. 

 

As for schools I worry teenagers may take the easy way out in pursuit of easy grades and free time. They'll be cheating themselves.   The world has no need for more high school essays, but that isn't why they're assigned. On the other hand student can learn about tasks these tools can help them with.  That's going to be more important in college.  I've been on two task forces and some schools are developing solid measures as well as the ability to change with time.  The guiding light is "are you teaching them how to think?"  

 

Clearly we're in a fascinating time.  I think we'll see some real carnage in the industry and use cases before signals we probably haven't thought about emerge.  But in the meantime we have beautiful art.   If you haven't seen WolfWalkers, you're in for a treat!

 

 

a legend

Adm Grace Hopper is of the heroes of early computing.  In addition to making fundamental contributions to software, she was one of the few who started to realize what technological change impacted people.  Recently some of the security folks - notably Bruce Schneier and Steve Bellovin - learned about a general talk she gave at the NSA in 1982.  They hounded the NSA to declassify and publish the talks and here they are!  They give a sense of what the world was like then and you can sense her brilliance.   This was a year before ARPANET cutover to TCP/IP, which many consider the real start of the Internet (although it goes back 13 years earlier in one form or another).  Perhaps there's something you can take away.  

Side story: About 12 years ago I found myself spitting into the wind trying to deal with climate deniers who lacked a technical background.  One of her papers gave a good description of what we now call sealioning. It was a good enough argument to convince me to move on and try other channels.  She also wrote about dealing with people like Trump who use bullshit asymmetry - what is known as Brandolini's law these days.  A very wise woman indeed!

 

 

 

playful active learning

A few months ago I had another in what has become a series of chats with a neurology professor.  Both of us are (well) over 60 and, given the two aging presidential candidates at the time,  the subject of preventing dementia came up.  He noted the standard things - exercise is a powerful tool, avoiding inflammation (you really don't want Covid), etc.  There's a genetic component, but that's more in dispute now as it breaks with twin studies - perhaps there's a bad habit component.  But actively learning is important.  

 

It seems scans of mentally sharp people with the sort of neural damage you see in dementia are able to continue to do well as they've created a web of alternate pathways - it's like the original idea for the Internet - route around the damage. Even the elderly can do this.  The trick is working hard at  something that really forces you learn.  Doing puzzles or reading is fine, but doesn't do the neural expansion. Learning a new language, how to build things, how to do something that's very confusing and hard at first creates a novel wiring overlay.

 

I've been known to plunge into new things for the fun of it.  I'm generally untalented, but love the experience of seeing some progress in an area I knew nothing about. My neurologist friend does some of the same. He likes playing board games so he's trying to create one.  It's given him a lot of respect for those that do it.  It's helpful to try something that might be fun as play is a powerful motivator.  If you like music, learn how to write something.  Try writing poetry.  The hardest part may be finding the time - but if you watch tv, perhaps it's the easiest. Turn on the curiosity.  I'd love to hear the projects some of you have.  It may be worse than what ChatGPT can do, but ChatGPT isn't giving you richer neural pathways. And there may be gains in your current level of creativity. 

 

Just now he sent a video that inspired him..  It's Xyla Foxlin.  She graduated from CWRU about five years ago and makes her living taking on unusual projects - (the rocket powered Christmas Tree is a personal favorite) Here she builds a mobius strip violin.  She admits she's not an instrument maker and goes on to use real luthiers for helping her with the bridge and a few other issues. It's a fun video!  And cheers to her for mentioning Hillary Hahn!

 

"exercise may be the single most potent medical intervention ever known:

a minipost - an important podcast episode that could have a large impact on your health

 

Eric Topol does cardiology and genomics at Scripps.  He's also an excellent communicators and was one of a half dozen voices during the pandemic offering translations of the rapidly evolving results from research and practice into language the general public can understand   He has interests that go beyond those areas an recently hosted Euan Ashley of Stanford on his podcast.  The title is provocative:  Exercise may be the single most potent medical intervention known. (audio and transcripts with links)  highly recommended.  The amount of sexual dimorphism in response is remarkable - but it's all heathly.

A couple of nits - noting I'm far from being an expert, but have read a fair amount in this area.. Ashley mentions the one minute five minute rule.  My understanding nothing supports this as a quantitative statement, but the qualitative direction is probably right.  Also Ashley mentions a company he's involved in that tries to optimize exercise for professional and serious amateur athletes. At the professional level much of this work is noise at this point - particularly the "AI" approach (and and LLM is clearly not the path), at the serious amateur level it may well be useful as anything is useful even though it may not be optimal. Research from the study Ashley talks about could lead to big improvements in this area.

What the podcast highlights is the release of data and some results  from a large body of research.  The research solidly confirms earlier work that points to how important exercise is as a medicine.

Here's the link to the data release site.  It has additional information - some aimed at the general public.

 

 

 

where goes art?

GOOSE -n

A bird that supplies quills for writing. These, by some occult process of nature, are penetrated and suffused with various degrees of the bird's intellectual energies and emotional character, so that when inked and drawn mechanically across paper by a person called an "author," there results a very fair and accurate transcript of the fowl's thought and feeling. The difference in geese, as discovered by this ingenious method, is considerable: many are found to have only trivial and insignificant powers, but some are seen to be very great geese indeed.

The Devil's Dictionary: Ambrose Bierce

 

I've had four conversations this past week about the direction of art given new technology and that catch-all buzzword term "AI"  Two of the people seemed convinced it was going to kill art and put artists out of business.  Their arguments were basically 'skill, particularly technical skill,  isn't needed with AI' and why would anyone look at a film or listen to music when they could just ask a computer to whip up something for them that they'd love.

 

These sort of arguments have been around since photography started becoming readily available around the time of the American Civil War.  Why would anyone learn art or hire an artist?   Film and radio were going to put theaters out of business, the theremin was going to kill the violin (!),  CG was going to kill the acting profession and so on.  And now we're hearing similar arguments. 

 

Capturing someone's voice without permission should be banned.. the same with their likeness, but LLMs are never going to write clear and compelling scripts given how they work.   They can and do write scripts and tens of thousands of GPT generated books have flooded Amazon, but clever and compelling, they're not.

 

There will be people who figure out how to use these technologies for art of one form or another of varying degrees of cleverness.  Unsurprisingly there's already a lot of effort in pornography, artificial voices to narrate audiobooks, ML driven image editing tools like Photoshop and even smartphones. 

 

One of the more interesting areas is in the storytelling part of console type computer games (arguably some of that is coming to phones, but perhaps the high end is more interesting in the short term).   There's interest in making NPC (non-player character) interactions more realistic.  These are the extra characters you can't control, but they're in the world space of the game.   I'm aware of a few sort of rough approaches    Down the road a bit NPCs may have more realistic interactions, but it's still much more structured that - say - telling the computer to just put together something without structure.  

 

Like the art of today I don't think you'll see much success in the long tail. Something special - often a combination of luck, talent and timing - will still be necessary.   Students studying the arts won't be wasting their time any more than they are today.  Actively doing art can fire your imagination and increase your ability to sense and think about the world.  Even a modest background can be useful in any number of fields.  And I suspect we'll see and hear a few wonderful expressions of human creativity along with a lot of garbage - just like today.

 

If you're interested in the technology and art of where all of this may be going the best resource I know is Lynn Cherny's Things I Think Are Awesome substack. I've known Lynn since the mid 90s.. she's always been deep in these areas.  $6 month/$50 year  

Also Amy Goodchild shows how she generates meaningful nonsense as an artform using a simple homebrew generative system.