The sorcerer’s apprentice

160 thoughts on “The sorcerer’s apprentice”

1. I can see teachers requiring students to write stuff in class. And then using software to compare that (vocabulary, phrasing, etc.) to what the student turns in as homework. The comparison software has already been developed for examining whether Famous Author A really wrote attributed work B, so it should be doable.

2. after deleting another 1000 words that became hopelessly muddled
   ChatGPT could have fixed it for you. 🙂

3. I can see teachers requiring students to write stuff in class. And then using software to compare that (vocabulary, phrasing, etc.) to what the student turns in as homework.
   A) That’s just more policing work that eats into a teacher’s already thinly spread teaching time. The constant erosion of actual teaching for other duties is a big factor driving teacher burnout.
   B) There’s a radical difference between the writing that an English Language Learner or Second Language student can produce in class and the work that they can produce with more time and revision. It’s already damn hard to tell the difference between the writing a 2L student can produce with revision and writing purchased from a paper writing service, and these cases are difficult to prove.
   C) The sample size one can get in in-class writing is not going to ever be enough to provide a robust base for this sort of textual analysis. Gotta have at least a few novels worth of writing to have any sort of certainty.
   Gotta decide the point at which trying to catch a cheat cuts into the actual teaching of the students engaging in good faith.
   That’s an ever moving target.
   There is, however, a pretty decent chance to catch the ChatGPT algorithm with another algorithm set to identify it as a writer, so that’s something.
   Ultimately, though, my goal is always to try to engage students with something that they have something they want to say about to a real audience, and then set them loose. If they are engaged and motivated to say something that means something to them there is less chance they are going to offload that onto a bot.
   And the real magic starts when you get them listening to each other carefully enough to start actually figuring each other out. All good writing is about actual connection.

4. I agree with nous about policing. Though a lot of the people talking about this here, both in the US and Japan, seem to think this is the logical fallback. It’s not, unless one of the two following things happen.
   1. Massively increase something, either the number of teachers, the amount of time teachers are expected to spend on this or the remuneration for teachers to make them willing to work harder.
   2. Pretend to increase the amount of scrutiny and policing
   The latter is pretty common in Japanese higher education, and a lot of the discussion about plagiarism, assignments and grades are often a pretense where the people who get caught are punished to the full limit while the bulk of the offenders slide. This isn’t limited to Japanese higher ed, a lot of enforcement depends on this. I feel like this is why it seems that Japanese commentators have not had the freak-out that Western teachers have had. A lot of my 1000 words was on how ChatGPT and other tools like this will not have the same kind of effect as they might have in English speaking countries.
   One reason for the lack of panic is that the bulk of Japanese students don’t and won’t see the value of being able to construct an essay. This is understandable, the English requirement for most Japanese students is meaningless, a required 2 years in university regardless of major.
   I also think that individual creativity is not the be-all, end-all in Japan that it is held up to be in the West. Japan has adopted stricter approaches to plagiarism and copying, but a lot of enforcement is as I mentioned above, people do it, and a few scapegoats are punished for everyone. And a lot of that stricter approach is Japan imitating the form, but not understanding the function.

5. I just don’t have, as it were, the bandwidth at the moment to read lj’s links (so sorry lj!), but I just want to say that as far as ChatGPT’s current incarnation is concerned, I found the examples that CharlesWT posted to be so hypnotically tedious and boring that (in the immortal words of Monty Python) it made me want to rip my own head off. Pity the poor teachers. At least bad students sometimes make funny mistakes to leaven the leaden.
   Apologies again for not engaging with the important pedagogical, administrative and political issues. I claim seasonal exemption.

6. I found the examples that CharlesWT posted to be so hypnotically tedious and boring that (in the immortal words of Monty Python) it made me want to rip my own head off.
   People who write political position papers should be very, very afraid.

7. I understand that reading through lengthy articles or papers can be time-consuming and may not always be feasible, especially if you have other pressing commitments. However, it’s important to note that not all research or writing is equally engaging or interesting to everyone, and that’s okay. It’s also important to keep in mind that academic research and writing serve a specific purpose and is often targeted toward a specific audience. As a large language model, I don’t have personal opinions or feelings, but I can understand why you might find the examples that CharlesWT posted to be tedious or boring. If you don’t have the bandwidth to read through the links that lj shared, you may want to consider bookmarking them for future reference or asking someone else to summarize the main points for you.

8. O brave new world, that has such chatbots in it.. (/sarcasm)

9. It’s also important to keep in mind that academic research and writing serve a specific purpose and is often targeted toward a specific audience.
   Disclaimer across the board: I don’t have the time or the bandwidth right now to immerse in this topic, fascinating as it is.
   But what I was most impressed with about the earlier ChatGPT comments was the fact that the sentence structure was orderly and correct. I’m skimming a lot, so I could be wrong, but I think the sentence quoted above is the first one where I would takeout my red pencil.
   Speculation: the correctness itself may contribute to the tediousness if it is achieved by having only a narrow range of acceptable sentence structures. The poems notwithstanding….

10. red pencil: takeout s/b take out

11. One alternate way to reduce the incentive to use software: go back to requiring writings to be hand written. Sure, someone could just copy what the software has generated. But the chances of miscopying, visibly, go up. Especially if the student wasn’t all that interested in the first place.
    Of course, it would mean that teaching handwriting would become more important. Heaven help anyone who has to read my usual handwriting — although my calligraphy is pretty good, if I do say so myself.

12. I find that my idea isn’t original (surprise!), even if it is independent.
    Here’s how teachers can foil ChatGPT: Handwritten essays by Markham Heid5. He also goes into some of the additional benefits for students.

13. One alternate way to reduce the incentive to use software: go back to requiring writings to be hand written.
    When I was in grad school most recently, getting my masters in public policy, all of the exams were mostly essay, written longhand, in blue books handed out at the beginning of the test. Most of the students complained of hand cramping by the end of the two-hour test periods. I never got around to asking what they would do if a student with arthritis said that they couldn’t write that much longhand, but assumed they had some sort of alternative.
    The first time I was in graduate school, the topology final was proving as many of seven statements as you could manage in three hours. A typical proof ran to a page or two of text including subject-specific notation. Given that it was topology, some of them required a quick sketch to illustrate a point (it was frequently said about topology that once you got the right picture in your head, the rest of the proof was easy). I don’t remember cramping being a problem, but I was much younger.

14. Topology cramps are the worst.

15. One alternate way to reduce the incentive to use software: go back to requiring writings to be hand written.
    My first reaction is to despair at having to decipher that many scrawls, especially as the word count expands. But that despair quickly gives way to much deeper philosophical and pedagogical questions.
    All of the suggested remedies for AI assisted writing remind me just how much non-teachers have internalized the idea that what we want from education is for individual students to demonstrate unassisted mastery over discreet tasks. We want to isolate students and compel them to demonstrate skill under pressure in formalized tasks.
    Aside from the tedium and difficulty of trying to decipher student handwriting, what strikes me about these handwritten assignments is the very artificial nature of the genre. Outside of school we rarely use handwritten notes for anything other than personal correspondences or working notes. Anything that is meant to engage a public audience is presented in a more legible, reproducible form.
    Requiring scholastic work to be handwritten and to be produced in isolation reinforces the artificiality of the writing. They are words being produced not for communication, but for satisfying some formal approximation of an ideal response with known parameters.
    Which is to say, not real writing.
    I actually think that more writing produced for classes should be collaborative and public facing, engaging with real audiences on real topics of collective significance.
    Again, I understand the need to evaluate individual progress and competence, but I’m not sure that trying to prioritize that as the capstone of a class is the right way to actually teach what is important.

16. An idea I had on my morning walk…
    A writing class in which the students take the prompt for the class and generate an exploratory draft of a paper using ChatGPT. The rest of the time is spent workshopping and revising that first draft to make it personally relevant and to adapt the writing to connect with a real audience of readers.
    The actual writing instruction here being mostly in the revision and in the tactical inventio, rather than in generating acceptable book reports of conventional wisdom on a topic.

17. I understand the need to evaluate individual progress and competence […]
    Of course. But are we not chasing ghosts here? ISTM, AI and the internet-information-age-always-online-computing society requires a markedly different skill set than a lot of what I had to learn as a kid. Memorization was such a big part of it, and I kinda feel like that’s not as important anymore when everyone has Google in their pocket. Cursive is a dying art and I don’t think penmanship is in the general curriculum anymore (of which I approve, as I was always bad at it – but even my block print is suffering these days – age and/or atrophy). I might argue that if we’re to give people the skills to function in modern society, we’re in need of a significant recalibration of what that means.
    I bailed on social media a while ago, but ChatGPT seems at least as lucid as a lotta stuff I saw on there – written by actual people that I know/knew. So… there’s that, as well.

18. One alternate way to reduce the incentive to use software: go back to requiring writings to be hand written.
    This has been floated in all the discussions that I have seen. Unfortunately, it clashes directly with another trend in education, which is to provide equitable access. My school advertises itself as a ‘barrier-free’ school because of the presence of a Department of Social Welfare that is quite forward looking in many ways. While my department has had a few handicapped students, the Social Welfare department has had some profoundly handicapped students with various conditions. The idea of just requiring them to handwrite their essays raises all sorts of issues.
    Also, a lot of teachers have adopted classroom practices that leverage technology. To ‘just’ have them go back to handwritten essays, yet keep all these other systems that rely on processed text would be problematic.
    nous point about the problematic need to evaluate individual progress and competence comes about because schools have a licensing function. You start poking at the notion of individual progress and competence and you start chipping away at the idea that the a school license is meaningful.
    I’ve actually started to use ChatGPT, my seminar students, who are getting to the point where they are finishing their papers, have been pointed to ChatGPT and have been asked to start using it to refine and improve their papers. This only works because they didn’t know about ChatGPT while they were writing it, and I have two students who have produced only a tiny amount of writing, so I’m not sure if it is kosher to point them to this. One student, who seems to have social issues, has hardly produced anything, having chosen a topic that was too difficult for him and resisting all attempts to get him to choose another. If I point him to ChatGPT, I feel like anything he produces will not be the result of any effort on his part.

19. The AI Content Detector detects AI generated text.
    QuillBot rephrases AI-generated text so that it’s less detectable.

20. I seem to have a comment in moderation.
    Fixed — wj

21. whatever happened to thinking?

22. Whatever happened to thinking?
    Oh, where did it go?
    It used to be so common
    But now it seems to be a no-show
    We used to sit and ponder
    And let our minds roam free
    But now it’s all about the screens
    And the endless scrolling spree
    Whatever happened to thinking?
    Oh, where did it flee?
    We used to have ideas and dreams
    But now it’s just reality TV
    We used to have discussions
    And share thoughts and opinions
    But now it’s just arguments
    And endless divisions
    Whatever happened to thinking?
    Oh, how did it slip away?
    We used to have conversations
    But now it’s just memes and slays
    So let’s bring back thinking
    And let our minds wander and roam
    Let’s have some deep conversations
    And leave the screens alone

23. whatever happened to thinking?
    My God, russell! Do you have any idea what that could lead to???
    People who think come up with new ideas. That leads to changes. Sometimes merely technological changes. But if you love your gas guzzler, and everybody starts making electric cars instead**? Oh, the horror! And social changes, any and all social changes, are worse.
    Not just that, people who think, especially young people, start questioning their parents’ deeply held beliefs. One of the few things left and right can agree on is that this cannot be tolerated.
    ** For obvious exonomic/business reasons. No government mandate required.

24. 135 prompts and counting that narrow ChatGPT’s responses to input.
    Awesome ChatGPT Prompts
    A list of tools and resources to use with ChatGPT
    Awesome ChatGPT

25. what I really miss are the crazy word salad emails that used to be auto-generated to defeat Bayesian spam filters.
    dadaist poetry at its best!

26. Need a version with a vocalizer that you can set to answer your phone and have fun with the spam callers.

27. Need a version with a vocalizer that you can set to answer your phone and have fun with the spam callers.
    Robots against (AI enhanced) robocalls!
    Reminds me of Ephraim Kishon’s recommendation to buy two chess computers so they could play against each other while oneself could do something useful in the meantime. [He was himself addicted to playing chess against the computer.]

28. I myself wonder where it gets its factual information, which seems to me to be terrible.
    I’ve asked it a few factual questions and the answers have been unimpressive – which is to say, often wrong.
    For example, I asked a few “Who is X” questions. Most were answered all right, but one, where X is a well-known academic, wikipedia entry and all, produced completely wrong information about X’s education.
    A similar thing seems to happen with poetry. It doesn’t scan. Look above for one example. ‘ve also seen it instructed to write specifically in iambic pentameter. No luck.
    I’m not so much criticizing here as wondering why these things happen. Is it a function of the design, inadequate training, or what?

29. Cursive vs. keyboarding is a much closer competition in ease & speed when writing an essay.
    BUT, when dealing with STEM topics, where equations and (simple) drawings are needed? Handwriting wins, hands down.
    Sure, sure, Microsoft Equation Editor. It’s utter shit. There are good options, but they are hard to use and slow. Writing by hand wins.

30. Most were answered all right, but one, where X is a well-known academic, wikipedia entry and all, produced completely wrong information about X’s education.
    Part of the problem may be time compression. Several decades or more of Internet facts are simultaneously true.

31. BUT, when dealing with STEM topics, where equations and (simple) drawings are needed? Handwriting wins, hands down.
    Sure, sure, Microsoft Equation Editor. It’s utter shit. There are good options, but they are hard to use and slow. Writing by hand wins.
    On the occasions when I teach Calc I, at some point I spend a lecture on notation (as a general “advanced math” thing). Newton may have been first, but his notation sucked outside of the specific applications where the only variable was time*. British refusal to adopt Leibniz’s notation ceded the field of analysis to the Continent for a century. eqn, or the stuff in Tex, or Equation Editor, or MathML, or a dozen other software tools are all attempts to produce symbology designed to be compact and easily hand-drawn on a keyboard. Almost without exception, people write down the hand-drawn version and then translate it. It’s kind of fascinating to watch Calc I students’ rendering and use of the integral symbol evolve over a few weeks.
    * Newton’s dot notation survives today in economics where, for the most part, all functions are functions of time.

32. Cursive vs. keyboarding is a much closer competition in ease & speed when writing an essay.
    I am in the process of breaking myself of cursive. While it still, in the words of my wife many years ago, “gives the appearance of great neatness”, it’s basically unreadable except by me. My mixed-case printing is much more legible despite being more compact. Timing while I wrote a few things, it’s not significantly slower than my cursive.
    I learned to play clarinet in fifth grade. I learned to type the summer after I was in sixth grade. There were a couple of years of piano lessons in there somewhere. I’ve been typing code on timeshare systems for more than 50 years. My handwriting is never going to come close to the speed I can do on a keyboard. (Well, maybe if the keyboard is a Teletype Model 33. Using one of those is the equivalent of weight-training for your fingers.)

33. Part of the problem may be time compression. Several decades or more of Internet facts are simultaneously true.
    Care to elaborate? I am talking about simple and readily ascertainable facts.

34. At different times, a fact may be true or false. For illustration, I ask ChatGPT for details on two separate people. Even though one of them hadn’t been editor-in-chief of a publication for more than ten years, it claimed that both of them were. It might not be drawing a distinction between what is true now and what was true in the past.

35. At different times, a fact may be true or false.
    Yes, but that doesn’t really apply to the case I was talking about. What it got wrong was where the individual in question had gone to undergraduate and graduate school, many years ago.
    In other cases it got the rules of a game wrong. Those rules haven’t changed in quite a while.

36. It will sometimes hallucinate “facts.” It has a real BS factor.

37. Me: Write a synopsis of “Harry Potter and the Sorcerer’s Stone.”
    ChatGPt: [Four paragraphs and 572 words.]
    Me: Make it terse.
    ChatGPt: Harry Potter is a famous wizard who discovers his powers and attends the wizarding school Hogwarts. He learns that the dark wizard Lord Voldemort, who killed his parents, is still alive and trying to regain his power. Harry and his friends Ron Weasley and Hermione Granger must stop a thief from stealing the Sorcerer’s Stone, a powerful magical object, and ultimately succeed in saving it.

38. Create a Wordle puzzle.

    H T A E L P S I
    R T I N U M O C
    A T S I L I G N
    E V I L R E T A
    N D E S I G N E

39. ChatGPT is the average entering freshman. It is capable of grammatically sound and mostly readable (if uninspiring) writing that is on-prompt, but it doesn’t know how to critically evaluate its information, has zero media literacy, and no concept of attribution.

40. Write pseudo-code for knitting.

    function knit(numStitches: int, numRows: int)
    for row in 1..numRows
    for stitch in 1..numStitches
    if stitch is even
    knit stitch
    else
    purl stitch
    end for
    end for
    end function

41. Using pseudo-code write the action logic for “The Old Man and the Sea.”

    old_man = Old_Man()
    sea = Sea()
    fish = Fish()
    while old_man.is_fishing():
    sea.update_conditions()
    if sea.is_rough():
    old_man.struggle()
    elif sea.is_calm():
    if fish.is_nearby():
    old_man.catch_fish(fish)
    else:
    old_man.wait()
    else:
    old_man.rest()
    if old_man.has_caught_fish():
    old_man.celebrate()
    else:
    old_man.mourn()
    

    Well…that’s enough.

42. Nowadays, MS Equation Editor accepts LaTeX input. I use it to write up notes when I want them to be legible for students.

43. The Wordle puzzle suggests that it doesn’t know which strings of letters are English and which aren’t. Or, at least as obviously, the rules of Wordle, which are so simple that you would think it would get at least the 5-letter-word part right.

44. Isn’t a Wordle puzzle simply any five-letter word?

45. Isn’t a Wordle puzzle simply any five-letter word?
    Technically, a Wordle puzzle in that sense is one of the words from the list of 2300 or so words on the solution list in the code for the game.
    The instruction “Create a Wordle puzzle” is ambiguous, at best. Does it mean to choose a solution? Does it mean to choose a solution and a possible sequence of attempts? How should color be represented in text-only input and output? Does it mean to create a Wordle-like puzzle (eg, one drawn from academic literature on optimal methods for solving Wordle-like games)? Would ChatGPT be likely to figure out that all words guessed must be drawn from an original list of 10,000 five-letter words (since expanded to about 12,000 but still not covering all five-letter English words), some of which are archaic, and not all of which are in the separate list of words allowed for solutions? For example, “ytost” is an allowed guess (archaic past tense of the verb toss), but will not be a NYTimes solution.

46. At a point when it wasn’t clear that the NYT was going to keep Wordle free, I programmed a wordle puzzle in Excel. (Yeah, I know.) I had it working fine except for the fact that I hadn’t created any libraries and I hadn’t yet handled giving clues for words with more than one instance of the same letter. I hadn’t thought through how I’d use it — probably just among the handful of friends I talk about wordle with, and we could trade of setting the solution word every day or something.
    Michael — your point about ytost inspires me to say that I have militantly ignored all the extras around Wordle. A friend and I do the puzzle every day, trade our puzzles when we’re done, bemoan the randomness of it, enjoy the randomness of it, and (in my case anyhow) thank Josh Wardle for making a puzzle that helps me do just one a day instead of what happens to me with sudoku and other such things….
    On BJ they’re always talking about all sorts of other puzzles that are out there for those of us who like Wordle. I just plug my ears to the siren call. (Though in truth, I’m not even tempted. Wordle is my fix and that’s the end of it.)

47. “extras around Wordle” = the kind of stuff that would have told me that ytost was acceptable…..

48. Nice try, “Michael”. Or should I call you by your real identity, ChatGPT?
    Because no human knows “ytost” is a word.

49. I thought it was a cheese:
    https://igourmet.com/blogs/gourmet-food-guide/gjetost-cheese-cheese-guide

50. Because no human knows “ytost” is a word.
    You badly underestimate how enthused some people get about the past. They typically don’t eschew the modern world entirely. But they really get into their hobby. Including the vocabulary.

51. Me, I was just scrolling through the section of words that start with y on the list of acceptable guesses and thought “No, that’s not a word…”
    Some poking around the internet has it come up in multiple epic poems (eg, Spenser’s The Shepheard’s Calender) by early Modern English writers who were striving to sound like an earlier era. Presumably there are some number of academics who would be aware of it.
    I have long said that anyone can become a world expert in something if they just narrow the subject sufficiently. In my case, I was one of the handful of experts on formal analysis of the feature interaction problem in telephone switching systems’ software. Becoming the leading expert on the use of “ytost” in literature is just sitting there waiting :^)

52. Technically, a Wordle puzzle in that sense is one of the words from the list of 2300 or so words on the solution list in the code for the game.
    OK. But that’s a completely arbitrary list, isn’t it?
    Also at some point, decades ago, when I worked at creating Y2K problems, I was told to never put things like lists of words (or, more often, tax rates, prices, item lists, etc. in the code itself.
    Has that changed?

53. As an aficionado of the Listener crossword (published in The Times on Saturdays, available also to online readers) I tend to be aware of quite a few obscure words, not excluding ytost.

54. Ok. Fine. I shoulda known the indwellers of ObWi would be the sort that tossed around “ytost” in casual conversation.
    Happy New Year, ya werd nerds!

55. I have long said that anyone can become a world expert in something if they just narrow the subject sufficiently.
    One letter words are my specialty.
    Happy New Year.

56. whatever happened to thinking?
    I’m gonna guess that’s at least semi-rhetorical but at face value the issue isn’t thinking. It’s how to test for thinking.
    Standardized proctored exams seem sufficient for STEM, broadly. For other pursuits, maybe orals (and essay)? Sort of a micro thesis-defense? We’ve got the Zoom thing down, mostly, so managing a schedule for that seems doable. Maybe not for auditorium-sized Intro classes, but anyone taking shortcuts at that level will be exposed soon enough.

57. Using Perl write code for the feature interaction problem in telephone switching systems’ software.

    use strict;
    use warnings;
    
    # Define a hash to store the list of features and their corresponding
    # implementation status (enabled or disabled)
    my %features = (
        'call waiting' => 'enabled',
        'call forwarding' => 'enabled',
        'three-way calling' => 'enabled',
        'voicemail' => 'enabled'
    );
    
    # Define a subroutine to check for feature interactions
    sub check_feature_interactions {
        my ($feature1, $feature2) = @_;
    
        # Check if both features are enabled
        if ($features{$feature1} eq 'enabled' && $features{$feature2} eq 'enabled') {
            print "Both $feature1 and $feature2 are enabled.\n";
    
            # Check for specific feature interactions
            if ($feature1 eq 'call waiting' && $feature2 eq 'call forwarding') {
                print "Warning: $feature1 and $feature2 may not work correctly together.\n";
            } elsif ($feature1 eq 'three-way calling' && $feature2 eq 'voicemail') {
                print "Warning: $feature1 and $feature2 may not work correctly together.\n";
            } else {
                print "No known interactions between $feature1 and $feature2.\n";
            }
        } else {
            print "Either $feature1 or $feature2 or both are disabled.\n";
        }
    }
    
    # Test the subroutine
    check_feature_interactions('call waiting', 'call forwarding');
    check_feature_interactions('three-way calling', 'voicemail');
    check_feature_interactions('call waiting', 'voicemail');

    Apparently, it will write code for just about anything.

58. But that’s a completely arbitrary list, isn’t it?
    Yes.
    Also at some point, decades ago, when I worked at creating Y2K problems, I was told to never put things like lists of words (or, more often, tax rates, prices, item lists, etc. in the code itself. Has that changed?
    That’s what I learned all those years ago. OTOH Wordle, both the original and the current NYTimes’ version, is a JavaScript applet running in a browser. Putting the fixed set of data in with the code has advantages. It saves a separate download for the data. It avoids potential problems with the browser, both present and future, not allowing code to access separately downloaded data for security purposes. It saves having to write even a simple parser for the data. It ensures that different players get the same data. At least for the NYTimes, it makes it difficult for anyone to use the game code in a more general way.

59. We may need to do an intervention for Charles.

60. 🙂

61. Stack Overflow had to temporally ban ChatGPT because it was overflowing with thousands of respondces people were posting.

62. So, you’re saying that Stack Overflow overflowed its stack…?

63. Using Perl write code for the feature interaction problem in telephone switching systems’ software.
    It’s a surprisingly good illustration of the basics of the problem: fixed set of features, fixed order of testing and/or code execution. If a new feature is added, then its interactions with the other features has to be added explicitly to the existing code. If you’re the telco and have three kinds of switch, you have to pay the vendors to do that job three times.
    We solved the technical problem. Our big suppliers hadn’t believed we could, so had signed contracts to do their side or pay big non-performance penalties. The week after we demoed the solution was one of the worst in my technical career: both of the big vendors called, said they had decided not to do their part — they could see a whole bunch of revenue disappearing — and asked if we wanted the penalty payments in cash or equipment credits.
    Well, I got a patent and a couple of conference papers out of inventing the key concept.

64. Technically, a Wordle puzzle in that sense is one of the words from the list of 2300 or so words on the solution list in the code for the game.
    OK. But that’s a completely arbitrary list, isn’t it?
    For some of us, the game has now become seeing how often one can come up with solution words which are almost guaranteed to not be in their list. At the beginning, they had an automatic message congratulating you on your vocabulary, but that is long gone, or even long ytost.

65. Et tu, GftNC?
    😉

66. Sorry, Pete. But happy new year all the same, to you and all on ObWi: here’s hoping we all (and our loved ones) have health and happiness in 2023.

67. So, you’re saying that Stack Overflow overflowed its stack…?
    Turns out the problem is that people are posting huge amounts of ChatGPT-generated problem solutions that haven’t been checked for correctness.
    The large language models are built in part from data scrapped from large swaths of the internet. People are putting huge amounts of LLM content on the internet. Some pundits think this is going to cause some kind of undesirable feedback.

68. data scrapped from large swaths of the internet.
    Is this not part of the problem? If the formative “learning” for AI is a bulk information dump…? Isn’t this recursive? Even with corrective inputs like Captcha and whatever other “human” guides(?) it’s learning from? The continuing bulk is still coming from a database continuously updated by unvetted information. So the algorithm weighs occurrence or frequency over time or… whatever metric? This isn’t unsolvable. But it ain’t there yet.
    Full disclosure, I gots no idea what I’m talking about.

69. Pete: “Full disclosure, I gots no idea what I’m talking about.”
    AHA! ChatGPT, caught you!

70. This isn’t unsolvable.
    Are you sure?

71. So, around here it has been New Year for 1 hour 1 minute by now.
    Correction: 1 hour and 2 minutes.

72. Whereas here, 2022 has another 8 hours to run. (Somehow we always get around to things later than the rest of the world. Well, except Hawaii.)

73. AHA! ChatGPT, caught you!
    You give me too much credit. 🙂
    Are you sure?
    No. But I kinda feel like the refinement of the algorithms will get close enough to be indistinguishable. Like, most of us listen to music converted to digital. Does it have the fidelity of analog? No. But can most of us make the distinction? Also, no.*
    *Yeah, ok. I’m talking to the “ytost” crowd. So yeah, y’all probably can tell the difference. Sticklers.

74. Technically, a Wordle puzzle in that sense is one of the words from the list of 2300 or so words on the solution list in the code for the game.
    Michael — I’ve been babysitting all day, so I’ve been pretty distracted. But as I think about it, it seems like the list you’d want to have handy is the list of 12,000 or so valid guesses, *not* necessarily the 2300-word solution list. Wouldn’t you only want/need the single word that is today’s solution on any given day?
    Wherever you kept the 2300-word list of solutions, you’d still have to feed the browser the information every day about which of those 2300 words is the right one today. So feed it an index to the word in the list, or feed it the word………either way you’ve got to do *something* each day to set up that day’s puzzle.
    Also, what are they going to do when they’ve exhausted the original list? (Sort of a rhetorical question….)
    Happy New Year — hoping for improvements. 😉

75. Also, would you assume that the solution word, however the browser is setup to recognize it, is encrypted?

76. Happy New Year to all. I’m going to go out and blow my stack.

77. Okay, i’m not going out or blowing my stack, so I’m back to Michael: I guess they could have programmed the date when every one of the 2300 possible solutions would be used — but there was an article recently introducing the woman they’ve made the Wordle “editor” — and it sounded like she’s picking the words day to day.

78. Sounds like a question JanieBot would ask. Be wary. ytost.

79. And there’s PeteBot trying to fool us again….

80. I’m gonna be hard-pressed to offer something stupider to prove my legitimacy if past comments are deprioritized.

81. … you’d still have to feed the browser the information every day about which of those 2300 words is the right one today. So feed it an index to the word in the list, or feed it the word………either way you’ve got to do *something* each day to set up that day’s puzzle.
    Shouldn’t be that hard to write a little piece of code that will (pseudo)randomly select a word for the day from the list.

82. Wordle editor article.

83. wj — true enough. Though you’d need to also make sure they weren’t re-used. (I quit playing the knock-off Worldle game because it re-used countries.) (I keep a list of words that have been used for Wordle solutions….)
    But see the article about the editor: they are purportedly NOT using either a pre-programmed list of word+date, or a random choice.

84. Happy New Year, luckily I never changed my clocks.

85. I keep a list of words that have been used for Wordle solutions
    We may need to do an intervention for Janie.

86. Heh. I have a whole spreadsheet devoted to Wordle studies. 😉

87. So you recognize that there’s a problem…

88. Keeps me off the streets…. Which is an unrecognized silver lining of a lot of hobbies and activities, no?

89. Yeah. The mean streets of… Belfast? Do dare, dear. What I wanna know is, how are you going to leverage your Mainey connections to get us in to the Lost Kitchen? I’ll pick up the tab and throw in a pair of questionably-woolen socks from Ocean State if you make it happen.

90. 25 or 30 years ago I had a lot of connections. The pathways tended to start in state government, but there are a lot of schmoozers there, so that’s not as useless as it sounds.
    I’m pretty much a hermit these days, but if I stumble across a Lost Kitchen pathway I’ll let you know. That would be a treat!

91. It’s in Freedom! Maine-wise, it’s practically next door!

92. But see the article about the editor: they are purportedly NOT using either a pre-programmed list of word+date, or a random choice.
    I opened up the JavaScript for the Wordle page at the NYT this morning. No list of solution words, but the list of acceptable guesses (including ytost) is still in it. The code has been minified, so I’m not going to look to see where the code gets the daily solution.

93. I know where it is….
    No time for stories now, but there’s a certain aspect of small town and rural life in Maine (probably everywhere) that means Freedom might as well be on the other side of the moon.
    Several years after my son was out of college, I ran into the dad of one of his high school friends. I asked about his kids and he gave me the rundown, including the fact that one of his sons was pretty far away — in China.
    China, Maine, that is.
    Freedom is even further away than China…..
    We could do a tour of the virtues: Freedom, Hope, Unity, Friendship…. I think there are a couple more….

94. I opened up the JavaScript for the Wordle page at the NYT this morning. No list of solution words, but the list of acceptable guesses (including ytost) is still in it.
    How do you access that? I don’t look at source much, but I thought that stuff was pulled from protected libraries anyway. No?

95. Liberty, Unity, Union, Harmony, Freedom, Friendship, Industry, Amity, Prospect . . . and Misery.
    We’ll stop before we get to Misery. 🙂

96. and Hope!
    (Wherever I got that list, they blew it. The quest continues.)
    (Gonna stop. I’m punchy, and I didn’t even go out last night.)

97. I know China Lake. I’m not one for “destination” dining, generally speaking. But I am intrigued by the rejuvenation of the town and that they repurposed the old mill as a power-generating plant. I’ve read Erin’s book and I recall walking by the place in Belfast years ago. She’s a self-promoter, certainly, but she does seem to be very connected and conscious about the local community. Would not have happened without the benificence of the Grassi family, but I like to think it’s generally a good map for the rejuvenation of smaller communities.

98. Taking optimism beyond the bounds of Maine:
    https://iiif.library.cofc.edu/iiif/2/36756/full/full/0/default.jpg
    I’ve a fondness for this one because it’s a postcard that my little brothers created, back when they were in the postcard business as Hold the Mustard Productions.

99. That’s a lot to take in, but Tranquility, CA is… Fresno?

100. Eureka, Texas is missing from the map…

101. Pete, it’s near the City of Fresno. In Fresno County.

102. Cool map, wj. 🙂

103. My siblings have artistic talent, no question. How they have managed to support themselves for decades is a bit of a mystery to me (the postcard business lasted for years, but never made all that much money), but they seem to have reached their 60s with food on the table, clothes on their backs, and a roof over their heads.

104. There are many tranquil places in the United States that could be considered peaceful and serene. Some possible options might include:
     The Great Smoky Mountains National Park, which is located in the Appalachian Mountains and known for its beautiful scenery and abundant wildlife.
     The Olympic National Park, which is located on the Olympic Peninsula in Washington and known for its old-growth forests, rugged coastline, and mountain ranges.
     The Yellowstone National Park, which is located in Wyoming, Montana, and Idaho and known for its geothermal features, such as geysers and hot springs.
     The Arches National Park, which is located in Utah and known for its striking red rock formations and sandstone arches.
     The Acadia National Park, which is located on the coast of Maine and known for its rocky coastline, wooded trails, and scenic drives.
     Ultimately, the most tranquil place in the United States will depend on an individual’s personal preferences and definition of tranquility.
     Nice try, Fresno. (courtesy of ChatGPT).

105. All of which have the common characteristic on NOT being towns.
     A town name of Tranquility must be seen as aspirational, rather than descriptive.

106. How do you access that? I don’t look at source much, but I thought that stuff was pulled from protected libraries anyway. No?
     For me on Firefox/Linux, at least, on the three-bar pulldown choose “more tools”, then “web developer tools”, then the “debugger” tab, then the “sources” tab. That gives access to a structured list of all of the JavaScript running for the current page. Selecting a script opens it in the editor section. Probably varies by browser.
     TTBOMK, there’s no way to stop the user from examining the JavaScript currently running in a standards-compliant browser.

107. I never drilled down that far, mostly because I didn’t think I could. Thank you!

108. TTBOMK, there’s no way to stop the user from examining the JavaScript currently running in a standards-compliant browser.
     I was wondering if it would be that easy for someone to just look up the day’s wordle solution. Although why anyone would want to is a mystery; if you don’t want to do the puzzle, don’t do it.

109. I was wondering if it would be that easy for someone to just look up the day’s wordle solution. Although why anyone would want to is a mystery; if you don’t want to do the puzzle, don’t do it.
     Hackers gonna hack. They don’t want to know today’s word for any reason connected to the game. They want to know because something about the question of how the Times’ code gets it struck their fancy. Why did people go to the trouble to prove that some questions about the game are NP-hard to answer, or even NP-complete?

110. As I’m sure you know, once upon a time and farther back than I’d like to think, you could just change “http” to “ftp” and get access to the whole public directory (and other things not as protected as they should’ve been).

111. Why did people go to the trouble to prove that some questions about the game are NP-hard to answer, or even NP-complete?
     You have traveled far beyond anything I know anything about, whether in terms of proofs or in terms of Wordle. I may masquerade as a technical person, but that rests on the fact that I had my first programming classes in 1967 (Algol), Fortran as an undergrad, and a systems programming class back at MIT when I was 30. Anything else I know I taught myself when I needed it. But it sounds like people are having fun with Wordle in many different ways. Yay!

112. But it sounds like people are having fun with Wordle in many different ways. Yay!
     You should see the academic literature on FreeCell :^)

113. I had a freecell addiction phase too….
     What about sudoku? I would assume that has generated a lot of study too. Even just thinking about how the puzzles are created makes my head spin.

114. Huh. I woulda thought Sudoku is fairly easy to reverse-engineer as that’s essentially what one does to solve it.

115. Pete — I guess I’m thinking of delinating levels of difficulty (e.g. easy, medium, hard, evil). Not sure I would have been able to grasp it even when I was younger, but now I’m old and lazy. And I don’t dare go near sudoku because that’s the one I can’t stop once I start, even worse than freecell.

116. But also, there are lots of pathways to a solution for any given puzzle, so if you reverse engineer, how do you know you’ve left enough clues to allow a solution? Just start emptying cells…hmmmm.

117. Wait ’til you hear about Minesweeper. 🙂

118. All of the games that require combinatorial approaches to solve/win (Wordle, Freecell, Sudoku as examples) attract a lot of interest. Figuring out how to solve them, or heuristics that come close, within time limits can be quite valuable. Every UPS truck leaving one of their facilities has a near-optimal pre-calculated route loaded that considers the delivery addresses for all of the packages on the truck and the time constrained pick-ups that truck has to make. UPS says they save at least tens of millions of dollars per year on fuel and overtime.
     I should do a post based on my lunchtime tech talk about the 7-11 combinatorial problem:

     On the way home from work I stopped at the corner convenience store to pick up four items. The checkout clerk did things on the register and told me “$7.11, please.”
     “That seems too much. How did you calculate that?” I asked.
     “I multiplied the four prices together.”
     “Aren’t you supposed to add the prices?”
     “Oh, right.” After a moment he said, “Still $7.11.”
     What were the prices of the four items?

     (If you start working on it, the obvious constraints apply. Prices are in dollars and cents, no fractional cents. $7.11 is an exact number using decimal arithmetic, no “pretty close”.)

119. Michael — do a post! Please!

120. I guess I’m thinking of delinating levels of difficulty (e.g. easy, medium, hard, evil)
     Well, yes? But there has to be “A Solution”. Or there could be multiple solutions. Consider a completely blank grid – it still takes some doing to make it all add up.
     Or am I even dumber than I thought?

121. Yeah, I’m just gonna wait for Charles to punch that into ChatGPT.
     🙂

122. “I multiplied the four prices together.”
     That’s shady.

123. Well, yes? But there has to be “A Solution”. Or there could be multiple solutions.
     Maybe a solution. I don’t know Sudoku enough to know if it’s possible to provide a partial layout such that no solutions are possible. Intuitively, it seems likely.
     The point, though, isn’t whether there’s a solution. It’s safe to assume one exists because the puzzle writers aren’t going to publish one with no solution. It’s a combinatorial problem, so will be solved by search. The point is whether one can organize (and limit) the search so that a human can do it in a reasonable amount of time.

124. It’s safe to assume one exists because the puzzle writers aren’t going to publish one with no solution.
     Yeah. Is it?

125. I have encountered sudokus that had more than 1 possible solution. It’s even possible, if only 4 open positions are left on the grid. I do not know, whether the problem of the minimal sudoku has already been solved, i.e. what the minimum number of positions filled is that still leads to a single solution.
     The minimum unsolvable sudoku is of course 2 positions filled that violate the basic rules, i.e. 2 identical numbers in one line.
     The easiest to grasp unsolvable one that does not violate the rules is filling in 8 times the same number in legitimate positions and then put another one at the only position left the last one would have to go to, i.e. blocking that position.

126. It’s easy to create a sudoku with more than one possible solution. If the completed grid is intended to look like this:
     . . . . . . . . .
     a . b . . . . . .
     . . . . . . . . .
     . . . . . . . . .
     . . . . . . . . .
     . . . . . . . . .
     . . . . . . . . .
     b . a . . . . . .
     . . . . . . . . .
     Then swapping a and b will give an equally valid solution. So one of those four digits has to be specified in the initial grid.

127. What I wonder is how many different ways a sudoku can be filled out if you consider the numbers to be arbitrary symbols, by which I mean if you equated, say, the top row left to right to 1 to 9 in order regardless of the actual number values and “decoded” the rest of the completed grid accordingly. And of them, how many are simply the same thing rotated or flipped over in whatever way?

128. What I wonder is how many different ways a sudoku can be filled out if you consider the numbers to be arbitrary symbols,
     I don’t understand what you mean, hsh. They’re already arbitrary symbols for sudoku purposes. You could just as well use abcdefghi. Or any arbitrarily chosen set of letters or symbols.

129. Whatever the set of 9 symbols is, the solution (grid all filled out) has to have one and only one instance of each symbol in each row and each column. Whether it’s digits or differently patterned chicken scratches doesn’t make any difference.

130. They’re already arbitrary symbols for sudoku purposes. You could just as well use abcdefghi. Or any arbitrarily chosen set of letters or symbols.
     What I mean is how many different patterns are possible, which is what I was getting at by suggesting that the top row of symbols left to right be equated to 1 through 9, even if they were numbers in some other order or whatever other set of symbols. (You could pick any column or row or 3×3 grid and do the same thing arbitrarily so long as you equate all the other symbols accordingly.)
     It’s an sort of an extension of the example of swapping a and b in Pro Bono’s comment. Let’s say you took a completed sudoku and, for example, changed all the ones to nines, all the twos to eights, and so on. The pattern would be the same even though the numbers would be in different cells. As I see it, that’s not really a different sudoku. It’s just a symbol swap.
     If you look at it that way, how many different sudokus are possible? (And is a mirror image of a given sudoku really a different sudoku?)

131. hsh — gotcha. Very interesting. Next lifetime….

132. The number is something over 5 billion, if you treat different number mappings, and row and column transpositions as not creating a new grid. At least 17 cells have to be given to make the solution unique. Wiki page.

133. The number is something over 5 billion, if you treat different number mappings, and row and column transpositions as not creating a new grid. At least 17 cells have to be given to make the solution unique. Wiki page.

134. By my logic, this is sudoku 1.0.
     123 456 789
     456 789 123
     789 123 456
     234 567 891
     567 891 234
     891 234 567
     345 678 912
     678 912 345
     912 345 678

135. ChatGPT says it appears to be complete with no errors. 🙂

136. ChatGPT is very wise. I will not challenge ChatGPT to a game of Tic-tac-toe.

137. Now if only ChatGPT could do italics. So it could say:
     It appears to be complete with no errors.
     Nuance is important for real communication.

138. I’ve just read four chapters of a Jack Reacher novel written by ChatGPT. Reads like bad fan fiction. But I can see that language models might someday be able to write publishable books.

139. But what would get published when everyone could read novels that were written just for them.

140. But I can see that language models might someday be able to write publishable books.
     But what would get published when everyone could read novels that were written just for them.
     There’s more to a good novel than correct sentences. I wouldn’t be surprised to see AI-written books along the lines of bad or even fair-to-middlin’ fanfic with properly grammatical sentences….
     But War and Peace? Middlemarch? Lord of the Rings? The Magic Mountain? Even a modern mystery series with a “series mystery” woven around the lives of the main continuing characters, with some emotional nuance…? Humor? Sadness?
     Hmmmm.

141. Hey, what about a sequel to Atlanta Nights?
     Really bad prose needs true experts. And it takes lots of discipline to not fall into old habits of good writing.

142. I’ve now read some of the links in the OP: my reaction is “What’s The Problem?”. Surely education in the humanities involves much more that assessing students’ ability to churn out formulaic, low-information essays.
     It’s long been the case that programs such as Mathematica can solve many mathematical problems. So much the better; one sets students questions which require genuine insight to answer.

143. GPT 4 is reportedly going to be released sometime this year. And reportedly going to be about five or more times larger than GPT 3.5. Will be interesting to see what the performance boost will be.

144. But what would get published when everyone could read novels that were written just for them.
     I’d be worried if I wrote porn. Yeesh…

145. It might be apocryphal, but supposedly there used to be humbugs in Egypt and Mesopotamia who complained about the advent of writing itself. It’s cheating if you don’t have to memorize stuff, after all, and the mental skills of future generations are bound to go into permanent decline as a result of this “papyrus” stuff.
     With that perspective in mind, I wonder if the implications of language models for teaching won’t ultimately be *that* different from other innovations, like electronic calculators, spell checkers, wikipedia, etc.
     To the extent that these models prove to be useful tools, the skills to use them *properly* will simply need to be taught like the use of those other tools. (And thus we progress one more step along the road to cyborgification.)
     Because I suspect they *will* prove to be useful tools. Leaving aside issues with intellectual property and information hygiene, software that generates coherent, on-point prose will obviously prove pretty indispensable in a lot of actual, workaday, real-world writing: manuals, research and policy papers, business proposals, legal documents, press releases, etc. A lot of that is pretty dry already, and I doubt there’s any harm in allowing an AI to at least take the first stab, at, say, writing up a tedious list of experimental results.
     Even when we’re trying to write something more creative or persuasive, it seems like there’s a lot of potential for these models to be used interactively. E.g., maybe it would help some people to use an AI to generate a rough first draft to overcome the ‘blank white page’ hurdle. Or just to flesh out a paragraph here and there — like content-aware fill for words.
     I’m assuming it’s only be a matter of time before tools like that are shipped with word processors. (If they aren’t already – I haven’t actually checked lately. Certainly Github’s controversial ‘Copilot’ is already doing something similar for computer code generation.)
     So soon the art of writing will partly lie in coming up with the right prompts, and editing to make sure the end product holds together coherently (including supporting material like quotes and citations). These are skills that will need to be taught.
     (The other side of this equation is that, eventually, a lot of prose might never be read by human eyes anyway. I suspect we’ll start to rely more and more on language models to analyze and summarize the material we’re consuming at least as much as we rely on other tools to generate it. Written material in many fields might end up more as a sort of machine interchange language than anything actually intended for human consumption.
     Which won’t be the end of the world. We’ll just be adding another layer of abstraction on top of the raw papyrus.)

146. Personally, I’ll be happy if they get an AI which will do “autocorrect” that actually corrects real typos, and doesn’t have us all cursing repeatedly. Which means it will have to know each of us individually, to have a clue as to our usual phrasing, vocabulary, etc.
     No sign of it here so far….

147. I can see using AI as a way to recompile texts. I won’t say “generate” because that’s really not what any texts do.
     This right here is the Poststructuralist precipice that I dare not cross. Beyond that lies unsustainable word count.
     I can easily imagine AI being used to create texts like The Naked Lunch or The Waste Land. They already seem to work off of a collage-like aesthetic, remixing a huge database of texts to conform to grammar and genre constraints. It’s a form of Searle’s Chinese Room. All that is missing for the texts to make the leap to something more is either a) curation by someone with good taste and sensibilities or b) tremendous patience winnowing down slush piles of crud.
     Either way, though, you are stuck with thee need for an aware subject in the mix, bucking thermodynamics to resist Sturgeon’s Law. You need insight to create (or recognize) texts that produce new insights.

148. @jack lecou — I’m thinking of how I’ve roll my eyes at the spate of articles I’ve seen suggesting that AI image creation is going to replace human photographers. Yes, I think, just like the invention of the camera meant that no one ever painted a picture again.

149. AIs could write bad fanfic. But could they ever achieve something like “The Eye of Argon”? (At least some of us here are old enough to remember when that burst on the scene.)

150. jack, what you are looking for is Plato’s myth of Theuth from his dialogue Phaedrus (Platon, Phaidros 274c–275b.)
     https://en.wikipedia.org/wiki/Phaedrus_(dialogue)#Discussion_of_rhetoric_and_writing_(257c%E2%80%93279c)

151. Here’s the full text and a bit of which what follows:

     Socrates:
     I can tell something I have heard of the ancients; but whether it is true, they only know. But if we ourselves should find it out, should we care any longer for human opinions?
     Phaedrus:
     A ridiculous question! But tell me what you say you have heard.
     Socrates:
     I heard, then, that at Naucratis, in Egypt, was one of the ancient gods of that country, the one whose sacred bird is called the ibis, and the name of the god himself was Theuth. He it was who [274d] invented numbers and arithmetic and geometry and astronomy, also draughts and dice, and, most important of all, letters. Now the king of all Egypt at that time was the god Thamus, who lived in the great city of the upper region, which the Greeks call the Egyptian Thebes, and they call the god himself Ammon. To him came Theuth to show his inventions, saying that they ought to be imparted to the other Egyptians. But Thamus asked what use there was in each, and as Theuth enumerated their uses, expressed praise or blame, according as he approved [274e] or disapproved. The story goes that Thamus said many things to Theuth in praise or blame of the various arts, which it would take too long to repeat; but when they came to the letters, “This invention, O king,” said Theuth, “will make the Egyptians wiser and will improve their memories; for it is an elixir of memory and wisdom that I have discovered.” But Thamus replied, “Most ingenious Theuth, one man has the ability to beget arts, but the ability to judge of their usefulness or harmfulness to their users belongs to another; [275a] and now you, who are the father of letters, have been led by your affection to ascribe to them a power the opposite of that which they really possess. For this invention will produce forgetfulness in the minds of those who learn to use it, because they will not practice their memory. Their trust in writing, produced by external characters which are no part of themselves, will discourage the use of their own memory within them. You have invented an elixir not of memory, but of reminding; and you offer your pupils the appearance of wisdom, not true wisdom, for they will read many things without instruction and will therefore seem [275b] to know many things, when they are for the most part ignorant and hard to get along with, since they are not wise, but only appear wise.
     Phaedrus:
     Socrates, you easily make up stories of Egypt or any country you please.
     Socrates:
     They used to say, my friend, that the words of the oak in the holy place of Zeus at Dodona were the first prophetic utterances. The people of that time, not being so wise as you young folks, were content in their simplicity to hear an oak [275c] or a rock, provided only it spoke the truth; but to you, perhaps, it makes a difference who the speaker is and where he comes from, for you do not consider only whether his words are true or not.
     Phaedrus:
     Your rebuke is just; and I think the Theban is right in what he says about letters.
     Socrates:
     He who thinks, then, that he has left behind him any art in writing, and he who receives it in the belief that anything in writing will be clear and certain, would be an utterly simple person, and in truth ignorant of the prophecy of Ammon, if he thinks [275d] written words are of any use except to remind him who knows the matter about which they are written.
     Phaedrus:
     Very true.
     Socrates:
     Writing, Phaedrus, has this strange quality, and is very like painting; for the creatures of painting stand like living beings, but if one asks them a question, they preserve a solemn silence. And so it is with written words; you might think they spoke as if they had intelligence, but if you question them, wishing to know about their sayings, they always say only one and the same thing. And every word, when [275e] once it is written, is bandied about, alike among those who understand and those who have no interest in it, and it knows not to whom to speak or not to speak; when ill-treated or unjustly reviled it always needs its father to help it; for it has no power to protect or help itself.
     Phaedrus:
     You are quite right about that, too.
     [276a] Socrates:
     Now tell me; is there not another kind of speech, or word, which shows itself to be the legitimate brother of this bastard one, both in the manner of its begetting and in its better and more powerful nature?
     Phaedrus:
     What is this word and how is it begotten, as you say?
     Socrates:
     The word which is written with intelligence in the mind of the learner, which is able to defend itself and knows to whom it should speak, and before whom to be silent.
     Phaedrus:
     You mean the living and breathing word of him who knows, of which the written word may justly be called the image.
     [276b] Socrates:
     Exactly. Now tell me this. Would a sensible husbandman, who has seeds which he cares for and which he wishes to bear fruit, plant them with serious purpose in the heat of summer in some garden of Adonis, and delight in seeing them appear in beauty in eight days, or would he do that sort of thing, when he did it at all, only in play and for amusement? Would he not, when he was in earnest, follow the rules of husbandry, plant his seeds in fitting ground, and be pleased when those which he had sowed reached their perfection in the eighth month?
     [276c]
     Phaedrus:
     Yes, Socrates, he would, as you say, act in that way when in earnest and in the other way only for amusement.
     Socrates:
     And shall we suppose that he who has knowledge of the just and the good and beautiful has less sense about his seeds than the husbandman?
     Phaedrus:
     By no means.
     Socrates:
     Then he will not, when in earnest, write them in ink, sowing them through a pen with words which cannot defend themselves by argument and cannot teach the truth effectually.
     Phaedrus:
     No, at least, probably not.
     [276d] Socrates:
     No. The gardens of letters he will, it seems, plant for amusement, and will write, when he writes, to treasure up reminders for himself, when he comes to the forgetfulness of old age, and for others who follow the same path, and he will be pleased when he sees them putting forth tender leaves. When others engage in other amusements, refreshing themselves with banquets and kindred entertainments, he will pass the time in such pleasures as I have suggested.
     [276e]
     Phaedrus:
     A noble pastime, Socrates, and a contrast to those base pleasures, the pastime of the man who can find amusement in discourse, telling stories about justice, and the other subjects of which you speak.
     Socrates:
     Yes, Phaedrus, so it is; but, in my opinion, serious discourse about them is far nobler, when one employs the dialectic method and plants and sows in a fitting soul intelligent words which are able to help themselves and him [277a] who planted them, which are not fruitless, but yield seed from which there spring up in other minds other words capable of continuing the process for ever, and which make their possessor happy, to the farthest possible limit of human happiness.
     Phaedrus:
     Yes, that is far nobler.

152. Yes, that’s it. I hadn’t remembered that was from an honest to goodness Socratic dialog.
     “[T]hey will read many things without instruction and will therefore seem to know many things, when they are for the most part ignorant and hard to get along with, since they are not wise, but only appear wise.”
     …eloquently recalls the “wikipedia expert” discussion from the other week. Time really is a flat circle, I guess.

153. Hartmut, thank you for that excerpt from Phaedrus. A perfect reference, under the circumstances.
     All that is missing for the texts to make the leap to something more is either a) curation by someone with good taste and sensibilities or b) tremendous patience winnowing down slush piles of crud.
     What a strange coincidence to read this in the week when I have been looking at the original typescript of The Waste Land, along with Ezra Pound’s alterations. I knew, of course, that he was Eliot’s miglior fabbro, but I had no idea of the truly enormous extent of his intervention, and his deletion of such huge amounts (although not of slush piles of crud). When I saw it, I almost felt that the work should be considered one of joint authorship, but people more qualified than I tell me that this is really the usual work of the good editor.
     And, as an aside, what a reminder too of how the appalling prejudices of the artist do not (and should not) contaminate the art.

154. FWIW, I have a really hard time taking Phaedrus at face value. I always wonder how much of Socrates expressed suspicion towards writing is Plato trying to make something more alive out of “dead conversation” and how much might be some deeper and more ironic relationship to Socrates’ suspicion. Specifically, I wonder how much of Plato’s portrayal of Socrates in this dialogue is meant to be read as faithful biography, and how much of it could have been a clearly fictionalized, ironic, or exaggerated caricature of Socrates.
     I think Plato is too smart to think that writing a dialogue gets one any closer to living language than does any other sort of fixed communicative medium. He had to have been self-aware enough to consider the disconnects and ambivalencies of his project, especially given the important role that drama played in Athenian life.

155. And back on the AI front, it strikes me that one of the good things that could come of the experiment I described earlier of having students start with an AI first draft of an essay would be that it would allow reviewers to be 100% honest and fully critical of the first draft in ways that one can never be when critiquing a writer’s own personal writing because the AI text is literally nothing personal. Would give you the opportunity to tear it apart without fear and then prioritize the revisions to get to a decent working draft that WAS personal.
     (And I write this, ironically, while procrastinating my revision of a bad writing prompt).

156. This past Summer I had occasion to write a short email that *had* to be in German, which I never knew terribly well, and had since been corrupted by bits of Italian, Japanese and French brain insertions.
     So: Google Translate for a first draft, then was able to dredge up enough to correct some obvious errors (mostly where English word choice gives ambiguous possibilities in German).
     Without the antique familiarity to make corrections, the pure-Google result would have been bad/awkward/embarrassing.
     And thus do we harness AI to our purposes.

157. Contrast the arguments being made by Socrates and Phaedrus in the following text..[Hartmut – 02:43 PM]
     ChatGPT: In this text, Socrates is making the argument that writing is not an effective way of imparting knowledge or wisdom, as it lacks the ability to engage in dialogue or clarify misunderstandings. He compares writing to painting, in that both appear to be able to communicate information but are in fact passive and unable to engage in a conversation or provide additional clarification. Socrates also suggests that writing can lead to a reliance on external sources of information and a decrease in the use of one’s own memory, leading to a false appearance of wisdom. Phaedrus, on the other hand, initially agrees with Socrates but later suggests that writing has some value as a means of preserving and passing on knowledge. Overall, Socrates is critical of the use of writing as a means of obtaining knowledge, while Phaedrus sees it as having some value but not being a substitute for direct communication and dialogue.

158. Professor Henry Jones: I wrote them down in my diary so that I wouldn’t *have* to remember.
     Ancient Greek philosophers had no concept of the scale that human knowledge would reach. I recall innumerable instances of “I don’t know that off the top of my head, but I know where to look it up” in my technical career. Heck, even “I don’t know that, but I know where to look up pointers to places where I can look it up” depending on the amount of detail involved.

159. Ancient Greek philosophers had no concept of the scale that human knowledge would reach.
     The last time one man knew everything is generally held to be at least 200 years ago. Exactly who that was is far less clear cut. And, of course, those designations all looked only at things “known in Europe” — probably why all the candidates are Europeans.

160. It was taught to me that education is mainly about where to find and how to understand and judge external info not to store raw data in one’s memory.

Comments are closed.