Yes, this is remarkably good!
I think it might have been simpler to give the AI a new Tinderbox document with just the Hints folder, ask it to do its readings, and then have it construct the primer in that file. But generating a new Tinderbox document is perfectly reasonable — Les Carr of Southampton used to do a lot of this.
Thanks. Given this, I can confirm there is a glitch if the feature. Tinderbox fails to detect/correctly parse a @ or # if they occur the start of the $Name string (position 0). As such, I won’t update aTbRef telling people to avoid placing these symbols at the beginning of the string, assuming this is a minor fix that can go in the next release.
I like the current approach as I get both the markdown files and the tinderbox file. I am sure if I give Cursor an empty Tinderbox file, may be with one note to start, it would manage as well. I learned a lot by this simple exercise. I also asked Cursor, what makes Tinderbox unique. I am sure you know. I renamed the markdown file to a txt file so that I am able to upload it here.
Thanks, this is very interesting!
I am surprised at the limited understanding of the range of Tinderbox’s views. My hunch is most compared apps have one/few views and those are of a core hierarchy/graph form and this confuses the AI. I do wonder if the AI understands that Map view isn’t a pure graph but a spatial hypertext. Hyperbolic view would be a better graph comparison. Quite why the Table view gets more prominence than the more powerful Attribute Browser view is also a head-scratcher.
The document is useful in that it shows some impressive synthesis but a less impressive understanding of detail. But if the need is high level overview it’s not too bad.
That’s an interesting competitive analysis. It gets a lot right about the ecosystem and the engineering issues.
Could you share that Python script?
After several months of working with different AIs, I have to agree with Sascha Fast. AI can be helpful if you are an expert in the subject. In other cases, it tends to be more of a hindrance because it tells you stories that are as true as those told in One Thousand and One Nights. They contain elements of truth, but much of it is pure fiction.
Another under-recognised fact is that, enquiring deeper, those making intentional use of AI and getting good results appear to (a) be using more than one AI, and (B) at a paid tier. Interesting, as I assume the investment money bet on AI means access costs are more likely to rise than fall, thus creating informational haves and have-nots. That is pertinent as writing guidance, other than from an entirely personal perspective, about Tinderbox+AI[brand] has a higher affordability bar than some other Tinderbox+OtherApp interactions.
At first, I was also a victim of advertising, but now I take a closer look. AI that can be switched off, preferably locally—why not?
I find that AI can be very helpful in areas in which one is inexpert, if used judiciously. A key is to use the AI to assist you in ways where good advice is helpful, and bad advice will be revealed promptly.
For example, answers to the question "What is the best recent book on ___” tend to be pretty good. If it recommends a book that doesn’t exist, you’ll find out right away and no harm will be done. If it recommends a bad book, well, that happens with real people, too.
Another good one is, "Look at this directory which has 1,700 text files. About how many lines of text are in those files?” Yes, I could write a program to do that, but Claude remembers C shell scripting better than I do. Again, if it just makes stuff up, you’re bound to know right away.
Another good topic: “Here is a phrase I don’t understand, in a language (or a computer language) I know imperfectly. Tell me about it.” This is a common problem for me: I’m reading along in French or in Swift and there’s something that makes no sense. I used to have to bug friends or random professors with questions; now, the easy ones can be handled by a machine.
My friend the composer sent me an interesting link to an oboist’s substack: ON TIME AND TECHNOLOGY. The writer observes of the Mozart Oboe Concerto that “the end is fucking brilliant”. I like this concerto, but I’ve never thought the end was especially notable. And do we mean the end of the first movement, or the end of the third? I asked Claude Haiku to explain, and was clearly at sea. Sonnet 4.5 has more insight, though I think it’s explaining why cadenzas work in the classical sonata movement, not really why this sonata is terrific. So, I guess I have to bother the composed after all.
Dear eastgate,
I would like to ask you a question
because I have some information that interests me.
What concert is Claude Haiku or Sonnet 4.5 searching for?
Oboe Concerto in B-flat major, K.deest
Oboe Concerto in C major, K.314/271k
Oboe Concerto in E-flat major, K.Anh.C 14.06
Oboe Concerto in F major, K.293/416f
Basically, there are no cadenzas created by Mozart himself.
D-major K.314, which is a Flute Concerto, is played by a famous performer.
There are Cadenzas (Taffanel, Andersen, Donjon, etc)
At Oboe Concerto you will find works by Henri Tomasi.
Oboe Concerto in C major, K.314/271k
Even Bärenreiter, which publishes the original version, does not seem to include Mozart’s cadenzas.
The most famous cadenza that I personally know may be the one by Heinz Holliger.
Wayback Machine Detailed information about Cadenza is written here.
The book also introduces some literature written by Mr. Raymond Meylan, an acquaintance mentioned in the article, but unfortunately I do not know the details of the contents.
There may also be an article about Mozart’s own cadenzas.
[http://idrs.colorado.edu/publications/journal/jnl24/cadenza.pdf\]
Oboe Concerto in F major, K.293/416f
Mozart fully orchestrated the movement up to the end of bar 50. From bar 51 to the first beat of bar 70 Mozart only wrote a single line (the oboe solo, plus the first violins in bars 62-63); everything past that was added by the completer.
A cadenza is also included as a suggestion, though performers should feel free to replace it with their own.
Usually when creating a cadenza, one respects the style of the period in which it was composed. Since it’s Mozart, it is naturally apply classical forms. As a result, the sonata form. It should be a basic musical idea.
Though , your phrases{ I think it’s explaining why cadenzas work in the classical sonata movement,] this is an interesting passage for me.
Yours, WAKAMATSU (from Japan)
Oh: another area in which AIs excel is in patiently explaining mathematical ideas. I sometimes get myself embrangled in math; both Gaudí and Hyperbolic views depend on math that I never studied formally.
In fact, I had to put Gaudí on the shelf for several months because I simply couldn’t work through a paper that described an algorithm I absolutely needed. Fortunately, about then a graduate student in mathematics needed a copy of Tinderbox and was willing to provide a few hours of Zoom tutoring. That got everything sorted out. But MIT grad students don’t always fall into your lap when you want them, and LLMs are very patient.