Daniel M. Russell's Blog, page 11

Interlude: Two SRS-style stories and the regretted loss of cache:

 It's been busy here at the SRS rancho... 

Search interlude. P/C DallE-3. 

... and so today is an interlude with a few stories you'll want to read.  

1.  First up, a story by Latif Nasser of RadioLab (the brilliant podcast) about his research quest.  In a longish Twitter/X thread, he tells the story of how he noticed a small, just slightly wrong detail in his young son's poster of the Solar System.  He noticed a moon orbiting Venus that was labeled "Zoozve."  A quick search told him that, as he thought, Venus doesn't have any moons.  He then "... googled “Zoozve” and got no results, literally zero results in English. Only results were in Czech and they were about zoos. Not what I was looking for." 

Close up of the original Solar System poster by Alex Foster.

After calling a NASA friend (Liz) and learning the same thing (yeah--no moons around Venus), he contacted the artist directly (Alex Foster), who said that he saw it on a list of moons... but he couldn't find that list any longer. 

Then, in a dramatic turn of events, Liz, the NASA friend called back saying "I think I figured it out."  Turns out there IS an object near Venus called "2002-VE"... could the artist have misread this as "Zoozve"?  

To be sure, 2002-VE isn't a moon, it's more of a wandering asteroid--a big rock that's currently orbiting Venus. Maybe you can call it a moonlet, or the more technical term, a quasi-satellite.    

Latif then called Brian Skiff at Lowell Observatory in Arizona--he's the guy who found it in 2002 as part of the LONEOS project--a very large-scale asteroid hunt for asteriods near Earth than could potentially smack into our planet causing a terminal event.

Along the way, Latif also found 2 astronomers who kept watching 2002-VE: Seppo Mikkola in Finland and Paul Wiegert in Canada. They told him that Zoozve is NOT a moon of Venus. But it’s also NOT really a moon of Venus. It’s both of Venus and not of Venus. 

A little more digging led Latif to a paper published in the Monthly Notices of the Royal Astronomical Society.   In their paper Asteroid 2002 VE68, a quasi-satellite of Venus, the authors point out that this moonlet has been in orbit around Venus for around 7,000 years, and will probably leave that orbit in 500 years (or so) until it's captured by another planet, or just wanders around the sun.  

As they write in the paper, "From the evolution of the orbit of this object, we conclude that it may have been a near-Earth asteroid, which, some 7000 yr ago, was injected into its present orbit by the action of the Earth."  

Latif then went on to learn that these quasi-satellites are complicated beasts, following extraordinarily complex orbits, and sometimes (as in the case of 2002-VE) moving their orbits between planets.  

SearchResearch Lessons:  

I love Latif's story here: it all started when he noticed a small detail that didn't fit in with what he knew ("there's no moon around Venus!"), so he started a research project to figure out what was going on.  I like that he failed in his Google search, but he kept at it, even contacting the artist and an astronomically learned friend.  

Liz, the NASA friend, had the key insight--maybe this was a misreading of something very similar!  Once you had the correct name for the thing, search gets a lot simpler.  

Bravo!  

(Postscript: Latif and Brian Skiff proposed that the quasi-satellite be called Zoozve... and the International Astronomical Union’s Working Group for Small Bodies Nomenclature (who keeps track of these things), ended up adopting the name, thereby making Alex Foster's poster correct retroactively.  Link to the paper that officially adopted the name.)  

Official announcement in the IAU's working group

2.  A story by Henk van Ess: Research into why planes sometimes travel faster than expected

On a recent flight, Friend-of-SRS Henk van Ess was puzzled: why did the flight's arrival time show as a full hour ahead of schedule?  What would cause that to happen?  

In his blog post at DigitalDigging.org (which I recommend to you), he gets to the bottom of this question through a conversation with ChatGPT and some sharp reasoning to determine that biggest factors for the faster times are:  (a) it's not tailwinds; (b) it's not the jet stream; (c) it's not having a light load with fewer passengers onboard.  

It IS primarily due to buffer times built into their schedule.  (But they did have a lack of headwinds, which also helped.) 

But his article points out that there's a big difference between ChatGPT 3.5 and ChatGPT 4 with the Data Analyst option turned on.  Your choice of LLMs really matter--understand which one you're using and be aware of the differences. 

SearchResearch Lessons:  In the end, Henk says that "... if you don't ask the right questions, you won't get the answers you need. Being clear and specific is key."  

We've talked about exactly this point many times here on SRS.  Understand your tools and be clear--very clear--about what you're asking.  

3. The loss of the cache:  

Most SRS Readers know that the cache: operator lets you pull up the last cached version of a web page.  (See the SRS discussion of cache: from 2018.)  

Alas, Google has deprecated this feature, so we can't use it any longer. (See Danny Sullivan's Twitter post about this.) 

In the meantime, I HIGHLY recommend that you get the browser Wayback Machine extension for your preferred browser.  Here's the list: 

     Chrome Wayback Browser Extension 

     Safari Wayback Browser Extension 

     Firefox Wayback Browser Extension 

     Edge Wayback Browser Extension 

Note that Bing still supports the cache: operator, so you don't really need it if you're an Edge user.  

FWIW, I suspect more features will be deprecated in the future, but I don't know what / when / or why.  It's just gonna happen as Google tries to save money and reduce the overall complexity of the search system from their side (and, presumably, invest those savings into Bard, or ).  

SearchResearch Lessons:  Things change... even operators.  Stay tuned because this will continue to happen.   

Back to our regular program tomorrow with an answer to the "How to summarize" Challenge from the previous post. 

Keep searching! 

SearchResearch Challenge (1/17/24): How can I get an AI to summarize a document?

We’ve got AI tools all over the place… 

… now, how can we use them effectively? 

An important piece of SearchResearch (or more generally, sensemaking) is finding relevant documents, reading them, and extracting the pieces of knowledge you need to get on with your work.  

I find that I spend roughly half of my research time just browsing, scanning, and reading—it’s what you need to do to understand a topic area enough to do real work. 

But what if we could accelerate the scanning and reading part?  What if we could take a long document and summarize it accurately?  

This is one of the great promises of the current crop of LLMs—they offer the ability to summarize long documents.  

Or do they?  Will they make the long document into a shorter, more focused work?  Or will that tool end up blurring everything into mush?  

The Challenge for this week is to explore how well an LLM can summarize a document in a way that’s useful for us SearchResearchers.  Here’s the Challenge for this week: 

1. How well does your favorite bit of AI technology do at summarizing long texts?  What system do you like for summarizing, and why?  

2. Test out your summarization method on two long-form texts.  First, let’s try that classic gothic tale, Frankenstein by Mary Shelley.  (Here’s a link to the full text that you can use.) How good is that summary?  

3. Second, let’s try your summarization method on a piece of text you might have read, but might have slightly forgotten—Chapter 10 of my book, The Joy of Search, link to the full-text of Chapter 10 here.  "When was oil first discovered in California?"  Did your summary method create a good/useful summary?  

Try using your fave summary method on these two works and let us know what you find.  We’re interested in practical ways of getting a good summary of long texts.  When you do this, think about what your goal is in getting the summary—is it just general interest?  Or do you have a particular question in mind?  When is a summary useful?  (And contrariwise, when does it not work well?)  Or, how do you know your summary is a good one?  

Let us know what you've learned in the comments section.  

Next week I’ll summarize what we’ve learned, and what techniques seem to work best for SearchResearchers.  

Keep searching! 

 

Special: How do you use LLMs in your SearchResearch

 Perhaps... 

... I was a little hasty in getting this week's SRS post out.  
Not long after I posted it, a few Regular SRS Readers wrote in with their methods of using LLMs. 

Regular Reader Unknown wrote in the comments that: 

I've found Bard useful for suggesting lines of enquiry. For instance, I asked "When did people in China first realise dolphins are not fish?" and it suggested some ancient Chinese texts which apparently discuss dolphins. I don't know if this info is accurate, but at least it's a good starting point. ChatGPT couldn't help with this query, and conventional searching didn't get me very far. I also asked Bard "Which art historians believe that Johannes Vermeer painted St Praxedis, and which are skeptical?" and it gave me two lists of names - again a good place to start.

I like this use of LLMs quite a bit--as a great research starter.  This is a great advanced use of an LLM resource. 

And I'd missed cmarlowe's earlier comment that: 

As a retired librarian I use LLMs to provide a basic understanding of new topic or to provide further insights into a topic I am researching and have some knowledge of. One of the features that helps me in researching is the conversations, because it requires me to think what else I should consider in my reference query. But always having that librarian mindset, I do check multiple search tools since often one finds different and helpful information to complete one’s research query and these can include general search engines or research sites. This also helps to avoid misinformation from LLM hallucination by comparing one to another. One unique LLM tool is Poe https://poe.com/ where you can search multiple AI sites and tools. 

Both Unknown and cmarlowe are using LLMs in a sophisticated way--they're using them as lead generators.  That is, the LLM gives them ideas about where to look and as useful resources.  

*

On the other hand, my friend Aditi sent me three examples where she used ChatGPT to answer questions that would have taken a lot longer with traditional search, and one where it's unclear that ANY search engine would have been able to do anything. This in a nice example of using LLMs to answer questions directly.  

Her description of her truly difficult search task was this: 

"[I wanted to] Find out why I remember an old Telugu nursery rhyme my grandma used to sing. I tried so hard with Google all these years and chatGPT helped me finally understand why I loved it and remembered it all these years." 

Here's their conversation of the first very difficult search task: 

Aditi:  I'm trying to figure out what an old Telugu nursery rhyme my grandma used to sing to me means. I'm not sure I remember the words right, but if I transcribe it to the best of my memory, might you be able to help me figure  it out? 

ChatGPT: Absolutely, I'd be happy to help. Please go ahead and transcribe the nursery rhyme as best as you remember it. 

Aditi: Ok here it is. I remember almost all of it clearly except for the fruit parts, which I've enclosed like (this?). I say fruit because it's about which bird eats which fruit I think? 

Gaagi, gaganaagi  

Ganganagi intilo paalupandi 

Aa pandu thecchi chilakakicchi 

Chilaka thinna pandu (cheedi?) pandu, Hamsa thinna pandu (arati?) pandu, [Baby's name] thinna pandu? Bangaaaara pandu!! 

ChatGPT:  Thank you for sharing the nursery rhyme. It's a charming Telugu rhyme about different birds and the fruits they eat, with a sweet twist at the end involving the baby's name. Here's a rough translation and interpretation: 

"Gaagi, gaganaagi 

Ganganagi intilo paalupandi" 

This part sets a playful and melodic tone, akin to calling everyone to gather around,  possibly imitating the sound of birds. 

"Aa pandu thecchi chilakakicchi Chilaka thinna pandu (cheedi?) pandu," 

It talks about bringing a fruit to a parrot (chilaka). The fruit could be "(cheedi?) pandu," where "cheedi" typically means a berry. The exact fruit is uncertain, but it's something the parrot enjoys. 

"Hamsa thinna pandu (arati?) pandu," 

This line is about a swan (hamsa) eating a  fruit, possibly "(arati?) pandu," which could refer to a banana in Telugu). 

"[Baby's name] thinna pandu? 

Bangaaaara pandu!!" 

This is the delightful twist where the baby's name is inserted, asking what fruit the baby eats, followed by "Bangaaaara pandu," meaning 'golden fruit', symbolizing that the child is precious and loved, like a rare and valuable fruit. 

It's a nursery rhyme that combines the joy of nature (birds and fruits) with the delight in a 

beloved child, turning a simple rhyme into a celebration of the child's presence. The uncertainty in the fruits adds a bit of mystery and room for imagination, which is often a charming aspect of traditional nursery rhymes. 

This is pretty impressive.  Bear in mind that Aditi transliterated the lyrics of the song she mostly remembered into a English-based transcription.  (Normally, Telugu is written in Telugu script, which looks like this: తెలుగు. That's "Telugu" in Telugu characters.)  And yet, based on her approximation of the Telugu lyrics, it somehow found the song and was able to find an English parallel translation.  

Both Aditi and I did regular searches to try and find such a resource, but we both failed.  There's a bit of knowledge buried deep in ChatGPT that includes Telugu nursery rhymes.  

And, in the spirit of double-checking all results (especially surprising results like this), Aditi checked in with her relatives in India and confirmed that what ChatGPT answered is, in fact, correct!  

*

In another search task, Aditi was searching for something to do with her young family on a cold day in Denver: 

Aditi:  What's something we can do as a family in Denver right now? It's cold so I was thinking something indoor and fun for a baby 

ChatGPT: Visiting the Denver Aquarium could be a delightful experience. Imagine it as a serene underwater world, a tapestry of vibrant marine life, captivating for both you and your young daughter. It's a warm, inviting space, away from the cold, where your baby can marvel at the wonders of aquatic life, creating a tapestry of early, colorful memories. 

Aditi: Nice! Any other ideas? 

ChatGPT: Consider the Denver Children's Museum. It's like a playground of learning and discovery, tailored for little explorers. Your baby would find a world scaled to her size, filled with interactive exhibits that engage the senses and ignite curiosity, while you could enjoy watching her discoveries and growth in such a playful environment. 

The responses are a little wordy, but they have the advantage of being high quality, and the disadvantage of being just one response.

Note that Aditi used ChatGPT's "custom instructions" feature to tell ChatGPT ahead of time who she was, about her family, and what kinds of responses she prefers.  (That's why the system knew that her family includes a baby.)  

This is a great way to tune ChatGPT to become more like the information resource that you'd like to use.  Aditi gave custom instructions saying that she has a graduate degree in information sciences, and that the replies it gives should be concise and accurate.  This has the effect of tuning the responses to be more like what you need. (I'll write more about this in the future--very handy.)  

SearchResearch Lessons 1. You can use an LLM as a smart collaborator that knows a lot about the world and can give useful direction.   This is probably the way I use LLMs most--as an extension of my brain.  (I recently asked Bard for [who is a famous writer on voodoo and hallucinogenic plants ] when I couldn't remember a particular writer's name.  It listed the top three authors in the field, with their names, and brief bios.  The first one is the author whose name I was trying to recall.
2. As we saw, even approximations of non-English texts can lead to great answers. I'm deeply impressed that a transliterated nursery rhyme was able to have ChatGPT find relevant information.  
3. Using LLM in a conversational mode can be incredibly useful.  Going back and forth with an LLM can focus the results to be more precisely what you seek.  
But, as always... double check your results. 

Answer: How do you use LLMs in your SearchResearch?

So.... 

P/C Dalle3. [an evocative picture of data, data tables, line charts, histograms]

It looks like many of us are using LLMs (Bard, ChatGPT, etc.) to ask SRS-style questions, especially ones that are a little more difficult to shape into a simple Google-style query.  

That's why I asked this week:  

1.  How have you found yourself using an LLM system (or more generally, any GenAI system) to help solve a real SearchResearch question that you've had?  

I've also gotten LLM help in finding answers to SRS questions that I couldn't figure out how to frame as a "regular" Google search.  Bard has answered a couple such questions for me (which I then fact-checked immediately!).  While the answers might have a few inaccuracies in the answer, the answers are often great lead-ins to regular SRS.  

For instance, I asked Pi (their home page) a question that has been in the news recently:  "What's the percentage of homeless people in the US?"  Here's what it told me: 

My first reaction was the obvious one--that can't possibly be correct!  18%???

The strangeness of the answer drove me to do a regular Google search for some data.  At the Housing and Urban Development website I found this: "The Department of Housing and Urban Development (HUD) counted around 582,000 Americans experiencing homelessness in 2022. That's about 18 per 10,000 people in the US, up about 2,000 people from 2020."  

Or, if you do the math, that's around 0.18% of the US population.  

When you look at multiple sites, you'll keep seeing that figure "18 per 10,000 people," which I could imagine an LLM rendering as "18%" in some twisted way.  Arggh!!  

Interestingly, when I did this on Bard and ChatGPT on the same day, they ALSO quoted 18% as the homeless rate. 

However... if you ask all of the LLMs this same question today (1 week after they told me it was 18%), they all get it right: the real percentage is 0.18%.  

That's great... but also worrying.  I'm glad they're in alignment with more official sources, but the underlying data didn't change--some training must have kicked in.  

You can't count on consistency with an LLM.  

* 

But... I find myself increasingly turning to LLMs to answer questions that really are difficult to frame as a search query.  Previously I wrote about finding "Other words that are like cottagecore" as an example of the kind of task that an LLM can do well.  They can take fairly abstract language ideas (e.g., "what words are like these words...<insert list>") and come up with something fairly useful.  This is a great capability that will let us take SearchResearch questions in new directions.  

You can ask follow-up questions without having to include everything you've found out thus far.  THAT is a pretty big advantage, especially for more complex tasks that take more than 2 or 3 queries to answer.  

* 

This past week I found a really interesting paper on our topic by some folks at Microsoft Research.  In their Arxiv paper, "Comparing Traditional and LLM-based Search for Consumer Choice: A Randomized Experiment" they compared ordinary search tasks where some people used a "traditional" search engine, while another set used an "LLM-based" search engine.  (The LLM was ChatGPT and was primed with an extensive prompt before people started their searching.) 

Intriguingly, they found a couple of interesting things: 

* people spent less time doing LLM search than trad search,

* they did fewer, but more complex, searches that trad search,

* on the whole, the two conditions were just about as accurate (but see below), 

* the searcher's perceptions of accuracy were pretty high, even when the answers weren't great.

They then did a nice intervention to see if they could color-code uncertain results in the LLMs output to see if it would help make everything more accurate.  The result?  Yes, significantly so.  (Although they still had to have a magic oracle know that the results were low-quality!)  

However, one of the Challenges they had their participants do was intentionally a hard question that they knew the LLM would give terrible answers. But it didn't matter much--people believe LLMs beyond what they should. 

That's not a huge surprise, but a good cautionary note for us.  You still need to check to make sure that the results you're getting actually jibe with reality. 

So... what do I use LLM search for these days?  I'm asking it a lot of knowledge summarization questions.  Things like "Give me an overview of how epigenetics works in humans.  Tell me assuming I have a college level degree of genetics knowledge."  

With this prompt, ChatGPT does a pretty good job.  Here's the opening paragraph of its response: 

Epigenetics is a fascinating field of study that explores the mechanisms by which gene expression is regulated without altering the underlying DNA sequence. It plays a crucial role in development, cellular differentiation, and the adaptation of organisms to their environment. Here's an overview of how epigenetics works in humans, tailored to someone with a college-level understanding of genetics... 

On the other hand, if you ask for an explanation at a 6th grade level of biology knowledge, you'll see: 

Alright, let's imagine our body is like a huge library. This library (our body) has lots of books (cells), and each book has the same set of instructions or recipes (DNA) for making you, well, you! But even though all the books have the same recipes, not all recipes are used in every book. Some recipes are for making eye color, some for hair color, and others for how tall you'll be. Now, epigenetics is like having special bookmarks or notes in those books that decide which recipes are going to be used or ignored....

These kinds of summaries are incredibly useful, and much more sophisticated than the abstracts / snippets beneath each web result on the SERP.  The ability to ask for more-or-less detail in the prompt is incredibly useful.  

* 

On the other hand... you still have to be careful about the simplest of things.  

If you ask Bard [ how many cups are in a gallon? ] it will reply: "There are 16 cups in a gallon. This applies to both US liquid gallons and imperial gallons, although they have slightly different volumes." 

Of course, that doesn't make sense. An imperial gallon is 1.2 US gallons, so they can't have the same number of cups!  This is a classic measurement blunder: there are, in fact, 16 Imperial cups in an Imperial gallon.  (For the record: There are 19.2 US cups in an Imperial gallon.)  As always, check your facts... and check your units!  (And for the record, ChatGPT explains this: "However, it's worth noting that the imperial cup and the US cup are not the same in terms of volume: 1 Imperial gallon = 16 Imperial cups,  1 Imperial cup = 10 Imperial fluid ounces.") 

SearchResearch Lessons

1. It's worth reiterating that you shouldn't assume that the output of an LLM is accurate.  Think of it as incredibly handy and useful, but NOT authoritative. Tattoo that on your hand if you need to, but never forget it. 

2. Always double check.  As we saw, LLMs will make bone-headed mistakes that sound good... so since you're not assuming that the output of an LLM is accurate, make sure you double-or-triple source everything. Do it now, more than ever.  

Keep searching!   

 

SearchResearch Challenge (1/10/24): How do you use LLMs in your SearchResearch?

 It's a New Year! 

P/C Dalle3. [an evocative picture of data, data tables, line charts, histograms]

Let's reflect on what SearchResearch is all about... 

I started this blog back in January, 2010 with About this blog--Why SearchReSearch?  That was 5,093 days ago.  

So far, there have been 1,374 posts, with 4.75M views and 13,700 comments.  We're running around 40K blog views / month, and that doesn't include the various syndicated versions of the blog.  If you include those, I'm Fermi Estimating the readership at 50K /month.  That's a decent number (roughly 1,660 people / day).   

As longtime SRS readers know, one of my goals was to use the blog as an effective prompt to write a book.  That book--The Joy of Search: A Google Insider's Guide to Going Beyond the Basics--came out in September 2019 and has done reasonably well. At least well enough to be translated into Korean and Chinese, as well as go into a paperback edition.  

As my Regular Readers also know, I'm working on another book ("Unanticipated Consequences") which I'm trying to finish up in the first part of this year, 2024.  

Writing a regular blog is a serious investment of time and energy, at least it is for me.  On average, I spend around 4-8 hours each week bringing you the most interesting SearchResearch tidbits I can find.  

But it's a big time commitment, and I'd like to use those hours to work on the new book.  

So I'm going to try an experiment--scaling back a bit, trying to do each blog post in two hours or less.  

To do that, I'm going to try to involve you Regular Readers a bit more in the search for insights.  

As we've discussed, the advent of Generative AI is promising to radically change the SearchResearch process.  It has the ability to give us new insights quickly (an example), and also to generate convincing-sounding nonsense with ease (an example).  

I know that many of you are using ChatGPT or Bard or Llama to answer questions, so I want to tap into the collective intelligence that SRS Regular Readers can bring to the discussion.  

I've asked this before in general (How can we use LLMs to search even better?) and for the specific case of medical searches (How might we best use LLMs for online medical research?).  But if there's one thing to know about LLMs and Generative AI in general, it's that this field is changing fast--really fast--fast as a flash rifling down the electric blue gun barrel into a sea of new-age synth-pop psychedelia.  

So... 

Our SearchResearch Challenge of the week is this is to revisit this: 

1.  How have you found yourself using an LLM system (or more generally, any GenAI system) to help solve a real SearchResearch question that you've had?  

 

I've told you before that I've used ChatGPT to help me with some data table manipulation. I've solved a few problems in a minute that would have taken me an hour or more to do with my regular programming skills.  Those are huge savings. 

I've also gotten LLM help in finding answers to SRS questions that I couldn't figure out how to frame as a "regular" Google search.  Bard has answered a couple such questions for me (which I then fact-checked immediately!).  

We all want to know what you've found to be useful.  What has worked for you in the past few months?  

Alternatively, what has NOT worked out for you?  

We're all ears here at SRS! 

Keep searching. 

P/C Duet AI (Google Slides) [kangaroo on top of piles of text listening ]

Answer: What makes a national dish?

 If fondue isn't a national dish,  

I'm fond of fondue.  
... what is it?  

Well, it IS a national dish, but as I mentioned last week, fondue was popularized as a Swiss national food by the Swiss Cheese Union (Schweizerische Käseunion) beginning in the 1930s as a way of using up the local excess of cheese.  The Swiss Cheese Union created pseudo-regional recipes as part of the "spiritual defence of Switzerland" before World War II to help out the cheese makers and the dairy industry.  (For more details, see this wonderful article about the cheese cartel that made fondue an international hit. Or, it you want to listen, this Planet Money podcast about the Swiss Cheese Union.) 

 This is all fascinating backstory to this week's Challenge.  What ELSE has been marketed as an authentic food?  I put the Challenge like this: 

1.  What other national foods have become popular as the result of intense marketing?  (I'm especially interested in foods that are presented as being "of the people," but are, in fact, commercial successes driven by clever advertising?)  Can you find one or two? 

I thought this might be a great question for the LLMs, and so I tried multiple attempts to get ChatGPT, Claude, and Bard to give me an answer.  

I'll spare you all of the prompts I created, but basically, I got very little.  With prompts like "what well-known national foods are actually the product of marketing campaigns?" I learned about marketing successes (e.g., Campbell's soups; Chiquita bananas; McDonalds hamburgers), but little about foods that are associated with particular countries (as fondue is with Switzerland).  

Bard: Of the suggestions Bard made: Poutine (Canada); Big Mac (US); and Hambúrguer Artesanal (Brazil).  Those are okay, I guess, but the marketing push behind each is a bit lackadaisical... nothing like the Swiss Cheese Union.  

ChatGPT: Slightly better: American fast food as a category (US); Fettuccine Alfredo (Italy); Sushi (Japan)... these are all popular foods, but none that were driven by clever advertising or marketing.  

One of the things that became clear to me was that I was a little ambiguous about what my SRS goal actually should be.  Was I looking for well-known national foods (like poutine or Big Macs), or was I looking for national foods that reached success because of an advertising/marketing campaign?  

As I often say here in SRS, you've got to be clear about what your research goal actually is so you'll know success when you see it.  

Naturally, it's okay to learn-as-you-go, but at some point you have to figure out what success will mean. 

I decided that I really wanted a well-known national food that became prominent internationally because of an intense marketing or advertising campaign.  

What did work?  Several friends wrote to me to suggest foods that they learned were market campaign-driven:  SPAM, chicken tikka masala, Vegemite, and even peanut butter were all suggested as advertising-driven successes of national foods.  

Reading up on these I found that: 

SPAM was given a big boost by the maker, Hormel, as the result of SPAM's ubiquity during World War II.  Advertising was a big component of pushing SPAM onto the global stage. (Massive distribution during World War II didn't hurt either, making SPAM incredibly popular in Oceania.)  

Chicken tikka masala, while immensely popular, doesn't seem to have been driven by advertising.  It's clearly a very British product that has become popular by word-of-mouth and news stories, even though it seems very Indian.  

Vegemite, the intensely-flavored Australian yeast extract spread, WAS intensely marketed at the product's initial production (in the 1920s) and led to vegemite becoming a distinctly Australian food.   

While peanut butter is often thought of as American, its invention is credited to three people with early patents on the production of modern peanut butter. (1) Marcellus Gilmore Edson of Montreal, Quebec, Canada, obtained the first patent for a method of producing peanut butter from roasted peanuts using heated surfaces in 1884.  (2) A businessman from St. Louis, George Bayle, produced and sold peanut butter in the form of a snack food in 1894. And  (3) John Harvey Kellogg (yes, THAT Kellogg) was issued a patent for a "Process of Producing Alimentary Products" in 1898 with peanuts, although he boiled the nuts rather than roasting them.  There's been a fair bit of marketing, but again, nothing quite as intense as the Swiss push on fondue.   

But how can we find all of these--and more!--by searching? 

My first regular Google search  was: 

     [ national dish food marketing campaign ] 

which led me to a fascinating book, "National Dish: Around the World in Search of Food, History, and the Meaning of Home" by Anya von Bremzen.  This is basically a compendium of the backstories of various foods, showing how sometimes national foods have a messy beginning.  For instance, she points out that Andalusian gazpacho were not just simple Spanish regional foods, but dishes historically cooked across the country. She points out that La Sección Femenina, the women’s branch of Spain’s fascist movement, identified some foods (like gazpacho) with specific regions, as part of its work to create “a sanitized, politically acceptable form of cultural diversity.”

I also found the Wikipedia page on National Dishes (have to admit that I wasn't expecting this page to exist--there must be a lot of foodies among Wikipedians). Scanning that list shows a lot of national dishes, some of which I recognized as very popular.  

I spot checked a bunch of them: Wiener schnitzel (Austria), empanadas (Bolivia), Peking duck (China), pot-au-feu (France) and about ten others. The nearly all of the national dishes I checked became national dishes over the course of great spans of time, without any particular interference or promotion.  They really are of the land and the people.  

However, two Japanese foods that are known world-wide qualify as marketing/advertising driven:  

Ramen is a massive international success that was driven in the US by the introduction of instant ramen noodles (beloved by college students), and by Chef David Chang's introduction of ramen at Momofuku in New York City in 2004, driven by a clever influencer campaign.   (Incidentally, Momofuku Ando was the name of the inventor of the instant ramen noodle.)  

Likewise, but to a lesser extent, sushi seems to have become an international success by marketing, although that effort was a bit hodge-podge with many small pushes from multiple players rather than one large company.  

I found those through friends (thanks, friends!) and by manual work from the National Dishes Wikipedia page. 

But then I had the thought, what if I limited my previous search to just Wikipedia? Like this: 

     [ site:Wikipedia.org  national dish food marketing campaign ] 

What would I find?  

Answer:  Several additional dishes and a bunch that you can look up on your own.  MY favorites from this search... 

Ploughman's Lunch (Britain, marketing campaign by the Cheese Bureau and the Milk Marketing Board, shades of the Schweizerische Käseunion!) 

Lamb (Australia, marketing by Meat and Livestock Australia, the MLA)

Kentucky Fried Chicken (USA, marketing by KFC, which scored a massive success in making KFC the default Christmas dinner in Japan)

Coca-Cola (USA, the Coke company has run one of the most successful marketing campaigns in history for over 100 years)

SearchResearch Lessons 

1.  LLMs sometimes just don't deliver.  I spent at least two hours trying to convince Bard and ChatGPT to give me something useful.  And yes, I tried all of the clever prompt engineering tricks I could think of, but nothing seemed to work.  Ah well... they're not the universal solution to all known problems. 

2. Friends!  I was happily impressed by the number of friends who had suggestions. Don't ever underestimate the value of a good social network.  I got text messages from some friends who'd seen the post, some comments on my other networks (e.g, LinkedIn or Facebook), and a couple of direct emails.  Ah, friends.  Good to have them as an extension to my brain.  

3. Wikipedia lists.  I should know by now that there is a list for almost everything.  National foods is not an exception to this rule.  Remember to look for "list of" something when you're searching for a category.  

4. Use site: over a large resource.  For instance, my use of my previous search query, when restricted to Wikipedia actually helped me find a few new suggestions.  Don't forget the power of limiting your search!  

As we close out the year, I'm going to take a couple of weeks off and start anew in 2024. I'll be moving back from Switzerland to California in the next few days and restart my life there.  It's been wonderful to explore Zürich and the Swiss countryside from a different perspective, that of someone who is new to the place--a new country to discover, what more could you ask for as a Christmas present.  That's what I got, and I hope you enjoyed learning a bit about the place as much as I did.  

See you in January, 2024.  Have a wonderful holiday season and New Year! 

Keep searching!  

SearchResearch Challenge (12/13/23): What makes a national dish?

 When in Switzerland, 

I'm fond of fondue.  
... you have to have fondue. 

Something I learned today…. 

Apparently, fondue has been around for a while, at least since 1875, when the first fondue recipe was published as a town-dweller's dish from the lowlands of the western, French-speaking parts of Switzerland. 

It was popularized as a Swiss national dish by the Swiss Cheese Union (Schweizerische Käseunion) beginning in the 1930s as a way of increasing local overproduction of cheese.  Too much cheese, too few cheese-eaters--the obvious fix is to figure out a way to increase local demand for local cheese.

And thus it came to pass that the Swiss Cheese Union created pseudo-regional recipes as part of the "spiritual defence of Switzerland" before World War II.  After wartime rationing ended, the Swiss Cheese Union continued its marketing campaign, sending fondue sets to military regiments and event organizers across Switzerland as a way to spread the message of social cheese eating. 

As a consequence, fondue is not just popular, but also now a symbol of Swiss unity. The marketing campaign successfully associates fondue with Swiss mountains, good times, and healthy outdoor winter sports.

Ever since those heady days, fondue has been promoted aggressively in Switzerland, with slogans like "La fondue crée la bonne humeur"  ('fondue creates a good mood') and, in Swiss German, "Fondue isch guet und git e gueti Luune" ('fondue is good and creates a good mood') – usually abbreviated as the tongue-twister "figugegl."  

Background: "History of Cheese Fondue" (Oct 29, 2009) – an interview with Isabelle Raboud-Schuele, by Gail Mangold-Vine.  Originally published in www.Automnales.ch (http://www.automnales.ch/) Raboud-Schuele is the curator at the Alimentarium Food Museum in Vevey.

 This is all fascinating backstory to this week's Challenge.  My illusion of fondue as the simple national dish of Switzerland has been shattered.  What I thought of as a regional, rustic dish of the commonfolk has been revealed to be the result of rather clever marketing.  

And that thought made me wonder: 

1.  What other national foods have become popular as the result of intense marketing?  (I'm especially interested in foods that are presented as being "of the people," but are, in fact, commercial successes driven by clever advertising?)  Can you find one or two? 

Of course, as always, we're deeply curious about HOW you found these other popular campaign-driven foods!  Let us know!  

Keep searching!  

Answer: What type of paintings are these? (Swiss Mystery #4)

 Curiouser and curiouser ... 

A knight striding with flag, bear, and sword. (Bern, Switzerland)

I'm officially a faculty member at the University of Zürich, and as such, you'd think I'd understand lots about Swiss culture.  But I'm a newbie, a recent (and temporary) immigrant.  I'm heading back to California at the end of the year.  
But one of the great things about traveling is the chance to see the world through new eyes.  Last week I posed a simple question: What are these paintings on buildings that I'm seeing everywhere in Switzerland.  
In the past week I've asked several Swiss folks what they'd call these things, and surprisingly, most of them said "I don't know... paintings?"  It's not something they've thought about much, they're just part of the background, they've always been there.  But to me, as an outsider, see the everyday and think it's extraordinary.

Here are a few examples: 

But to me, these stand out and are interesting.  Since I'm a curious fellow, I had to ask the Swiss Mystery for this week... 

1.  Is there a name for this particularly Swiss kind of artwork-on-the-walls?  Is there a particular name of the style in which most of them are drawn?  

How to start?  

I know that paintings on walls are often called frescos, so: 

     [ Swiss frescos ] 

actually didn't work all that well. I found a couple new ones, but the new paintings I found were mostly interior frescos (a la the Sistine Chapel in Rome, probably the world's best known fresco).  So it's pretty clear they're not called that.  My next search: 

     [ Switzerland painting on exterior of buildings ] 

worked a bit better.  This led me to find that many people used the word "facade" with these paintings.  My next query: 

     [ Switzerland facade painting ] 

returned results that were pretty good.  There were many more "Swiss exterior paintings" in the Images collection, but I still don't have a collective noun for these things.  Besides, "facade" means "the principal front of a building, that faces on to a street or open space," which is nice, but it doesn't describe the paintings per se, but just where they're located.  

On the other hand, this results page points to some great collections of these paintings.  One of these links is to the wonderful paintings on buildings in Schaffhausen: 

   https://www.swiss-spectator.ch/fassad... 

I put the full text of the link there because I noticed something odd. See that term in the URL "fassadenmalerei"?  I don't know it... could it be a clue? 

If you do a search on: 

    [ fassadenmalerei ] 

you'll find a lot of sponsored links--ads for companies that paint the facades of buildings.  The German word "fassaden" means "facades" and "malerei" means "painter," specifically of decorative paintings.  So this makes sense--of course there are companies that will paint your facades, usually in elaborate geometric color schemes.

I looked at a few of their sites just to get a sense for the kind of work they do.  Do any of them do paintings of the shown above?  

The fourth site I looked at does something they call "wand-und-fassadenmalerei."   Since "wand" means "wall," maybe we're getting a little closer. Their demo images include some of the type we seek.  

While reading a few of these sites, I found another word that seems on our trail: "Lüftlmalerei."  A quick search for: 

     [ Lüftlmalerei ] 

takes us to the German Wikipedia entry, Lüftlmalerei, which tells us that

"... Lüftlmalerei (also spelled Lüftelmalerei ) refers to the art form of facade painting native to small towns and rural areas in southern Germany and Austria, especially in Upper Bavaria (Werdenfelser Land) and in Tyrol."  

Well, that's interesting.  

The region of Tyrol is in the far western part of Austria, bordering eastern Switzerland.  Southern Germany is to the north and east of Switzerland, so it's definitely in the area. 

Interestingly, when I asked my Swiss friends about this word, Lüftlmalerei, they mostly said they understood it as "air painting," and that it sounded somewhat familiar, but they also all said "it sounds like an Austrian word."  (Apparently that second L, lüftlmalerei, gives it that uniquely Austrian sound.)  

Looking at lüftlmalerei images shows us that this seems like the right word: 

Some SRS Regular Readers wrote in (on one of our parallel syndications) about this Challenge.  

Regular Reader Bill Janssen just knew that the word was probably “Wandmalerei."  As we saw, that's certainly part of the answer.  

Franz Miltner (who works in Ingelheim am Rhen) says  “Not sure if there is a different term in Switzerland, but it's Lüftlmalerei in southern Germany and Austria."  

Perhaps most notably, Regular Reader SpiritualLadder points out that: 

You might not like this, but I literally put in your exact question with the image into ChatGPT and this is what I got:

(ChatGPT reply): "The artwork depicted in the image you've provided is reminiscent of a mural. In Switzerland, such murals are often associated with a long tradition of public art and can be reflective of various styles, from classical frescoes to more modern interpretations.

Specifically, in Switzerland and parts of Germany, there is a tradition known as "Lüftlmalerei," a style of mural painting that is particularly found in the alpine regions. This style often features pastoral scenes, historical figures, and local legends, and is characterized by its use of vibrant colors and intricate details to create a sense of three-dimensional space on flat surfaces."

When you google Lüftlmalerei, you find that it is present in Germany and Austria, so it sounds like it might not be the right answer, but they do look similar to the ones you posted...so it feels right. (Maybe it has a different Swiss name.) If it *is* the right answer, that would explain why it's hard to find!

I felt like ChatGPT (with vision) *might* be able to handle this kind of task well since it might have been trained on images of this style. The first time I prompted it with a different wording it wasn't as helpful, but putting in your exact question got me this answer!

To SpiritualLadder's point, it doesn't bother me at all that ChatGPT was able to provide such a good answer--this is a real improvement! 
As I've noted before, the LLMs are still subject to hallucinations, so you still need to check everything.  But if this accelerates the search process for some Challenges, I'm all for it.  
SearchResearch Lessons
The traditional way I did it illustrated some insider techniques. 
1. Don't ignore the URLs.  They sometimes contain clues that will help you find the result you're looking for. 
2. Read related sites while looking for terms and concepts you don't know.  Reading related websites is how I learned about the idea of Lüftlmalerei, quickly doing a couple of searches to validate that this is, in fact, the correct term.  
3. Don't ignore the LLMs.  As SpiritualLadder points out, the LLMs can sometimes do a great job of pointing out things that you might not have considered.  Be sure to validate the results they give you, but they're getting better every day.  Consider checking AND confirming!  

Keep searching! 

SearchResearch Challenge (11/29/23): What type of paintings are these? (Swiss Mystery #4)

 As I wander over hill and through dale, through alpine passes and mountain streams... 

A knight striding with flag, bear, and sword. (Bern, Switzerland)
... I'm seeing a kind of exterior artwork on the outside walls of many Swiss buildings.  This is such a common thing that surely there must be a specific name for these kinds of paintings that appear on the buildings.  Sometimes they're old, sometimes relatively new.  Some fancy, but most are stylized in a way that I cannot describe easily--they're often noble, or heroic, frequently bucolic, sometimes civic in design, almost always in a kind of flattened, non-realistic rendering. 

The Swiss Mystery for this week is this... 

1.  Is there a name for this particularly Swiss kind of artwork-on-the-walls?  Is there a particular name of the style in which most of them are drawn?  

I admit that I do not know the answer yet, but I feel as though there MUST be a rather specific term for these images on the walls of greater Switzerland.  

Here are a few examples: 

What do you call these?  More importantly, what do the LOCALS call this kind of artwork?  

I'd like to learn more about this artistic tradition, but don't know quite where to start.  Any ideas from the SearchResearch crowd? 

When you find the answer(s), let us all know HOW you found them!  Let notes in the comments field. 

Keep searching! 

Answer: How does it work? Checking your assumptions?

We all make assumptions... 

... it's a normal thing to do. But when we're doing online research, it's good to check yourself.  This week's Challenge is a little story about why...   

My assumption was that these common kitchen appliances had the same mechanism for knowing when the toast / water / rice is at the right temperature or level of doneness. 

Then I checked--and had a big surprise.  Really?    

Today's SearchResearch Challenge is simple: 

1.  So... how DO each of these devices know when the toast / water / rice is ready?  

2. (extra credit) What other devices do you believe you understand, but when you checked, you learned that you actually didn't understand?  Does anything spring to mind?  Any surprises?  

Searching for the answer isn't hard:  

     [ how does an electric tea kettle know when
         the water is boiling ] 

     [ how does a rice cooker know when
         the rice is done ] 

     [ how does a toaster know when
         the toast is done ] 

Let's talk about these one-by-one.  

Electric tea kettle: Do this query and you'll find a lot of explanatory text telling you that a ring-shaped bimetallic strip changes shape as it warms up, and since it's slightly buckled, when it deforms just enough, it snaps into a different shape, mechanically flipping the heater switch to off when the water hits the boiling point of 100C (212F). Very simple, very clever.  But how does it know?

The underside of a tea kettle with the bimetallic ring outlined with a red dashed line.  

That bimetallic ring is actually a fairly smart little invention--the tang coming out of the ring lies in the plane of the ring at room temperature, but as the ring starts to bend under the heating of the kettle, it suddenly snaps to a different shape with the tang pointing up.  This is called a bistable device, a gizmo that can be in one of two different physical configurations. Of course, the bimetallic strip's clever trick is that it will flip into the "hot" state when the temperature gets high enough, then flip back to the cool state after the heat is off.  That rapid flip to the other bistable position is what mechanically flips the power switch back off. 

The best example of this I could find is this YouTube short showing a tea kettle thermostat that flips back and forth rapidly when warmed up.   (Video link.) 

And... I thought that was that.  

But as I was writing this up, I ran across a video about electric tea kettles on Steve Mould's YouTube channel that showed me something I'd completely missed:  the steam tube! 

(Steve Mould video short on steam tubes in tea kettles.)  

It turns out (as Steve Mould points out), that the thermostat is actually at the bottom of the tube that is open to the top of the vessel.  As the kettle heats up, eventually the steam in the kettle is forced down the tube where it directly heats up the thermostat, which flips at around 95 C (203 F).  

Why is this important?  Because water boils at different temperatures at different altitudes, so a thermostat that is preset to 100C won't switch off at high altitudes.  In Denver, which is at 1609 meters (5280 feet), water boils at 95°C, or 203°F.  So if the thermostat only tripped at 100°C, it wouldn't work!  

At this point I got really curious about this somewhat sophisticated design, so I looked up a few electric tea kettle patents, where I learned that this is well-known by tea kettle designers.  To quote one such patent (US4357520, from 1979):   

"Alternatively a steam tube or passage, which communicates with a steam or vapour aperture in the upper wall of the container, may be run down the outside of the container. Such a tube or passage may be concealed within or behind a handle structure of the container..."  

I thought that was it. NOW I understood how they worked. 

But as I was writing this up, on a lark, I searched for: 

     [ electric tea kettle diagram ] 

hoping to find a cross-section of a kettle showing the steam tube.  

But NO!  I was again surprised to learn that there is a second thermostat, a "boil dry" thermostat that kicks in if/when the first thermostat fails to detect any steam.  This would happen if the kettle was switched on without any water.  No water, no steam, no steam pressure, no thermostat clicking off. 

I found this because I saw this diagram in the search results: 

Electric tea kettle circuit diagram. (P/C from Karisimby's blog)
I know enough about electronics to realize that this was NOT what I had in my head.  The "steam thermostat" was what I expected.  What I did not expect was a second "Boil dry" thermostat.  

A bit more searching revealed that such a second thermostat is set at a much higher temperature and is attached directly to the kettle wall.  So if the kettle is dry and heating up, the steam thermostat won't kick in, but the boil dry thermostat will open the circuit and stop the heating from running away (and possibly causing a meltdown).  

Amazing what you'll learn if you keep poking.  (If you want to learn more, I recommend this wonderful explanation of tea kettles.)  

Rice cooker: Again, not a difficult query, but the answer surprised me!  I'd assumed that there was a simple bimetallic thermostat in the rice cooker as well--in fact, I'd assumed that it worked in much the same way, probably with an identical part inside. 

But I quickly learned that most rice cookers (especially older ones) use a very clever magnet that turns off the heating element when the temperature goes above 100°C, or 223°F. 

Basically, there are two heaters in a rice cooker--one that keeps the rice warm and a second heater that boils the water, much like the tea kettle. 

However, I learned that rice cookers have a little magnet that holds the switch closed to close the main circuit to the heating element.  That circuit heats up the water until it boils, after it boils, the temperature of the magnent will start to rise beyond 100°C, or 223°F. When that happens, the magnet loses it's magnetism, allowing the circuit breaker to pop open, stopping the heating.  

The amazing trick here is that the magnet is designed so that it's Curie point (that is, the temperature at which it loses its magnetism) is at 100°C!

This is simpler to watch a video than to explain in text.  So I recommend this video: Nice explanation of how the magnetic cutoff switch works on a rice cooker. 

The bimetallic strip thermostat works because different metals expand at different rates, causing it to change shape. 
I didn't dream that using a magnet's Curie temperature would be used in the simplest of all rice cookers.  It's really not obvious that this is would be used to turn off the heat.  Incredibly smart use of some sophisticated physics.  
Oddly, I can't find any evidence that a rice cooker has a boil dry mechanism!  (I assume they do.  If you find out, let me know in the comments below.)  

To be sure, there are other mechanisms out there for telling when the rice is done--neural fuzzy logic rice cookers with complicated sensing systems.  But I love this elegant application of magnets!  

Toaster:  The big problem with toasters is that there are so many of them.  But for the most part, simple toasters are just timing mechanisms... turn the dial, get the level of brown you'd like.  It's really up to you learn which dial setting corresponds to the level of toasted that you like.  
But there are a LOT of different toasters out there--some with fancy thermocouples to determine temperature, some with just plain old spring-loaded timers that count down the seconds, and some with complex electromagnets to hold down the carriage (that carries the bread/toast on its journey).  
The old-fashioned toaster, though, is just a timer... nothing more than that.  

Other devices I don't understand? 
2. (extra credit) What other devices do you believe you understand, but when you checked, you learned that you actually didn't understand?  Does anything spring to mind?  Any surprises?  

There are LOTS of things I don't understand, but will look into once I get the chance.  I want to learn how the magic happens. Among them, 
1. iPhone--how it knows how far I've walked or run.  I know it counts steps, and I know it has a GPS system in it, so I assume it uses the GPS for distance traveled and counts steps via the on-board accelerometers, but does it combine those information streams in some way?  Don't know.  
2. Refrigerators--I don't understand the physics of them in detail, but I do know that they have a single cooling unit which then cools the frozen foods at one temperature, but the cool part of the fridge is handled in a different way.  Again, how does it know?  
3. Sauté  release--Sometimes when I cook (say) a piece of meat, it will come out of the pan easily--it "releases" with ease.  But sometimes, it sticks and makes a mess.  Is it purely a function of temperature, the amount of oil, or the material in use?  I need to understand this!  
The bigger point here is that all of these phenomena are understandable with a little observation and a dash of SRS.  

SearchResearch Lessons
1. Expect surprises and be ready for them.  Initially I thought that the rice cooker story was going to be straightforward.  So it was a surprise when the story got more complicated.  I discovered that when I noticed that something I read from the SERP didn't fit in with what I already knew.  (And this happened to me twice!)  
2. Use multiple document types to get a full perspective.  Sometimes a regular web search works just fine, but for complicated searches like the rice cooker, I used Image search, YouTube searches, and even Patent searches to get multiple perspectives.  I ALSO looked at multiple sources for each device, partly because some of the documents are simplified versions.  (For instance, not all rice cooker stories mention either the steam tube OR the dry boil cutoff thermostat.)  
3. Be aware when what you read doesn't match what you think you know.  Again, this happens multiple times for every SRS session--I learn things, and then learn something different.  When what you read doesn't line up with what you've already read, take that as a signal that there's more to learn!  
And keep searching!