jamestown (part 2) Cities
2025-04-07 18:21:15

This seems to be "evolving" into a dictionary. Somewhat by accident; the machine wrote a command-line frontend for me (without my asking) that looks a lot like a dictionary.

----

When are two definitions the same? ๐Ÿ”ฅ when they are not different. and they are different if the part-of-speech differs, or the Chinese translation differs. and some other times.

----

These three seem to be the same.

Word: eternal

Rank: 2547

Definitions:

[1] Existing or continuing without end; everlasting.

Confidence: 0.95

Part of speech: adjective

Lemma: eternal

Chinese: ๆฐธๆ’

Subtype: duration

Notes: Often used in philosophical or religious contexts to describe the nature of existence or the divine.

Examples:

- The concept of eternal life is central to many religions.

[2] Lasting or existing forever; without beginning or end.

Confidence: 0.90

Part of speech: adjective

Lemma: eternal

Chinese: ๆฐธๆ’

Subtype: duration

Notes: Can refer to abstract ideas, such as love or truth, that are considered timeless.

Examples:

- Her love for him felt eternal, transcending time and space.

[3] Relating to or being a part of eternity; timeless.

Confidence: 0.85

Part of speech: adjective

Lemma: eternal

Chinese: ๆฐธๆ’

Subtype: duration

Notes: Often used in literary or poetic contexts to evoke a sense of infinity.

Examples:

- The stars in the night sky seemed to whisper eternal secrets.

On the other hand, these three might be different:

Word: transform

Rank: 2646

Definitions:

[1] To change in form, appearance, or structure; to metamorphose.

Confidence: 0.95

Part of speech: verb

Lemma: transform

Chinese: ๅ˜ๅฝข

Subtype: change

Notes: Often used in contexts involving significant change or conversion.

Examples:

- The caterpillar will transform into a butterfly.

[2] To cause to change in character or condition; to convert.

Confidence: 0.90

Part of speech: verb

Lemma: transform

Chinese: ่ฝฌๅ˜

Subtype: change

Notes: Commonly used in contexts such as technology, business, or personal development.

Examples:

- The new software will transform the way we manage our projects.

[3] In mathematics, to change the coordinates of a point or the representation of a function.

Confidence: 0.85

Part of speech: verb

Lemma: transform

Chinese: ๅ˜ๆข

Subtype: change

Notes: Used in various branches of mathematics, including geometry and calculus.

Examples:

- We can transform the equation into a simpler form.

----

and these are definitely different:

[1] A piece of furniture with a flat top and one or more legs, used for placing items on or for working at.

Confidence: 1.00

Part of speech: noun

Lemma: table

Chinese: ๆกŒๅญ

Subtype: small_movable_object

Notes: Commonly used in homes and offices.

Examples:

- We gathered around the dining table for dinner.

[2] A systematic arrangement of data, usually in rows and columns, for easy reference.

Confidence: 1.00

Part of speech: noun

Lemma: table

Chinese: ่กจๆ ผ

Subtype: small_movable_object

Notes: Often used in academic and professional contexts.

Examples:

- The research paper included a table summarizing the results of the experiment.

[3] To postpone consideration of a motion or proposal in a meeting or legislative assembly.

Confidence: 0.90

Part of speech: verb

Lemma: table

Chinese: ๆ็ฝฎ

Subtype: other

Notes: Usage can vary by region; in some places, it means to bring a motion forward for discussion.

Examples:

- The committee decided to table the discussion until next week.

jamestown (part 1) Cities
2025-04-06 20:39:52

โš™๏ธ Jamestown, North Dakota, is located along I-94 in the eastern half of the state.

Today's focus is on "word frequency".

----

I started with two corpuses: one of 19th century literature (from Project Gutenberg), one of 20th century "sci-fi" literature. I got a rough word-rank for each, and combined them โš™๏ธ using the harmonic mean to get a combined word-list.

While many high-frequency function words such as the, and, and of maintain consistent rankings, others like said, her, she, and me show substantial divergence, suggesting notable stylistic or thematic shifts between the two periods and genres.

----

It is also a word-list at all. Some of the notes:

  • The word "whale" shows up a lot more in the 19th century corpus. This is because one of the books is Moby Dick.
  • I am hoping to run an exhaustive listing of a few attributes. These include:
  • polysemy. โš™๏ธ I am less concerned with words like get which have so many meanings as-to be indefinable, but instead words like saw (็œ‹ or ้”ฏๅญ) or face (้ขๅ‘ or ่„ธ)
  • by lemma. "went" (108th) v. "go" (80th).
  • by part-of-speech. defined as "what the LLMs define as part-of-speech".
  • a "second-level" of word-type details.
travel news Misc
2025-04-06 03:58:03

seen on Reddit: what US states border each other but don't have an Interstate Highway that crosses that border? on Reddit, the examples given were Kansas-Nebraska, Nebraska-South Dakota, South Dakota-Montana, with the diagonal "four corners" states being a special case.

Claude ... attempted to write out each list (borders and interstate connectivity) directly, and then use JavaScript to find the missing entries. It found Michigan-Wisconsin, and Idaho-Nevada.

ChatGPT ... kept giving incorrect answers, and trying to weasel its way into being right. New Hampshire-Maine (connected by I-95), West Virginia-Kentucky (connected by I-64), Vermont-Massachusetts (connected by I-91, which does not enter New Hampshire), Arizona-Utah (connected by I-15), Delaware-Pennsylvania (connected by I-95). โš™๏ธ https://chatgpt.com/share/67f1fb2f-8d18-8007-9e51-c3d68a162a3a ... I think I broke ChatGPT by telling it to use the colors

----

Looking at a map, Iowa-Wisconsin, Missouri-Kentucky, Idaho-Wyoming, and Oklahoma-New Mexico were missed.

glamorgan (part 3) Orinoco
2025-04-02 16:31:03

other than the chaos of the law, what might be next?

  • it would be nice to have the torrent server for some of these Git repos. ๐Ÿ”ฅ I have home bandwidth, I might as well use it. ๐Ÿ’ก this would also involve some "metadata management"
  • i mentioned an "adaptive difficulty vocabulary quiz" app. DuoLingo for first-language study. or, in other words, a smart dictionary ๐Ÿ”ฅ maybe that is "grenora"?
  • "camera-ready output" for LLM prompts. the first example being "school worksheets".
  • a tool that organizes documents. ๐Ÿ’ก step 1: read the document and generate metadata. step 2: sort and title documents. step 3: profit.

----

i mention an "analytics stack". but that would be "install an existing one", not "write one myself".

----

for greenland, the only outstanding task is "make Claude work in the unified client". this will also do things for atacama.

glamorgan (part 2) Orinoco
2025-04-01 23:24:30

I find myself reading the actual transcript, and wondering how Mr. Yalowitz actually kept a straight face.

JUSTICE GORSUCH: So, if I've got it right, you want us to adopt the fundamental fairness language from our Fourteenth Amendment jurisprudence but give it different content in the Fifth Amendment. Is that fair?

MR. YALOWITZ: I don't think you have to adopt the fundamental fairness --

JUSTICE GORSUCH: No. But, to get your five votes, that's -- you're willing to do that?

MR. YALOWITZ: I would be willing --

JUSTICE GORSUCH: Yeah.

MR. YALOWITZ: -- to do that.

JUSTICE GORSUCH: Okay. And if we did that --

MR. YALOWITZ: I'm not going to lie.

So far I don't see how anyone could think any of the justices would support Mr. Yalowitz, who is arguing to overturn the case.

โš”๏ธ well, actually, there may be another lawyer with better arguments.

----

JUSTICE THOMAS: Didn't the Justice Department take a different position in the 1980s, that the PLO did not have constitutional rights?

MR. KNEEDLER: Well, you say a different position. We're not taking a position here. But, in those -- those cases were --

JUSTICE THOMAS: Well, it's different from the non-position.

----

Mr. Berger makes an admirable attempt at a very silly position he is forced to take:

JUSTICE THOMAS: Mr. Berger, I know this is not the center of your argument, but could you just explain how PLO and PA are persons within the meaning of the Fifth and Fourteenth Amendment?

MR. BERGER: Yes. Absolutely, Justice Thomas. And I would respectfully draw the

Court's attention back to the February 2018 amicus brief filed by the United States in an earlier round of this case, in which the government's non-position was considerably more of a position and basically said there's only one type of person that this Court's precedent has recognized is excluded from the Due Process Clause and that's sovereigns. And, because the PA and the PLO are not recognized as sovereign by the United States, they are by default persons entitled to due process protection. ... But, ultimately, it's binary, right? If -- if -- you're either a person for purposes of the Due Process Clause and entitled to due process protections, or you're a sovereign state and you're entitled to the protections of the Foreign Sovereign Immunities Act, but there's no no-man's-land where you're neither a sovereign state nor a person.

Alito seems the only one to buy into the theories:

JUSTICE ALITO: What exactly is the unfairness in -- in this case?

MR. BERGER: In --

JUSTICE ALITO: It's the -- it's too burdensome to litigate this in New York where the PA and the PLO conduct some activities?

MR. BERGER: Well, so --

JUSTICE ALITO: What's the unfairness?

๐Ÿ”ฅ it is somewhat amazing that only one of the Supreme Court justices is a partisan hack.

----

It is hard to imagine anyone other than Alito supporting the position that the court should find jurisdiction.

glamorgan (part 1) Orinoco
2025-04-01 22:41:32

what is Orinoco?

candidate A: business software

candidate B: education software

candidate C: a list of philosophical prefects

candidate D: :guild_sign_1:

candidate E: a law blog ...

----

the law doth not extend that far

In that sense, what is happening in DC is predictable. DC is the center of an excess of laws, and DJT is a maximal chaos agent. ๐Ÿ’ก this is why the news is bad. people see this and think "He's working for Russia?" No. But, he is "chaos" and Russia is, in some way, "chaos" ... if you play just a bit stupid, you end up with "working for Russia". so, that's what the media says.

----

today, at the Supreme Court: Fuld v. Palestine Liberation Organization โš™๏ธ https://www.scotusblog.com/2025/04/justices-likely-to-allow-damages-suit-against-palestinian-authority-to-go-forward-in-u-s-courts/

At issue the case is the concept of personal jurisdiction โ€“ whether courts have the power to hear a case against certain defendants. In 2019, Congress passed the Promoting Security and Justice for Victims of Terrorism Act to give federal courts personal jurisdiction over cases by terrorism victims abroad. The law provides that the PLO and the PA โ€œshall be deemed to have consented to personal jurisdictionโ€ in any civil case brought under the Anti-Terrorism Act, no matter when the act of international terrorism occurred, if they make payments to the families of terrorists who injured or killed a U.S. citizen or engaged in any activities within the United States. The law carves out a few narrow exceptions to the latter criterion โ€“ for example, for conduct relating to official United Nations business or meeting with government officials.

The U.S. Court of Appeals for the 2nd Circuit ruled that the dispute now before the court, brought by U.S. citizens injured in terrorist attacks in Israel, as well as the families of U.S. citizens killed in such attacks, could not move forward. It reasoned that the Palestine Liberation Organization, which is the official representative of the Palestinian people for foreign affairs, and the Palestinian Authority, the governing body for parts of the West Bank and Gaza Strip, had not agreed to have U.S. courts exercise jurisdiction over them. Nor, the court of appeals added, is the conduct that the PSJVTA targets the kind of conduct from which it would be fair to infer consent.

----

๐Ÿ’ก it is a matter of an excess of law.

the United States Congress cannot "deem consent" for a foreign entity to be subject to US Courts.

this is a matter of fact.

whether they choose to do so anyway is an open question.

๐Ÿ’ก there is an excess of law. it does not extend that far. now it shall diminish.

----

There does not seem to be any coverage other than SCOTUS Blog. Which says that "Justices likely to allow damages suit".

I see it differently. Clarence Thomas is not going to go for this. Neither will John Roberts. 5-4 that PLO is not subject to the judicial Power of the United States regarding its actions in the territory of Mandatory Palestine (1922). No amount of legal mumbo-jumbo can allow the US Congress to "deem consent" by a foreign party, when those parties who clearly deny it.

minot (part 5) Cities
2025-03-31 22:05:35

Another round of "chainsaw coding" with Claude. ๐Ÿ”ฅ the generators now use generators! ๐Ÿ’ก Previously, Claude's approach was to just make one question without history. But we do want history. Use all the canned questions from text files (once), then generate files from local data (without repeating), then ask the LLM.

It mostly works. The LLM generation is slightly flaky. It gave both "sad" and "melancholy" as candidate opposites for "happy". It is including the pinyin for Chinese translations some of the time.

But at least it is (mostly) using the correct interfaces now. ๐Ÿ’ก there are still some useless parameters. The "tags" for questions are meaningless. The difficulty is arbitrary. The evaluation criteria are often silly.

----

The various "general knowledge" benchmarks should be quick to write ๐Ÿ”ฅ tomorrow. Right now I see six categories for the initial tests:

  • History
  • Geography
  • Chemistry
  • Biology
  • Sports ๐Ÿ”ฅ American athletes, mostly
  • Music ๐Ÿ’ก 20th century English-language music, mostly

These will be "easy" questions. Or, at least, multiple-choice.

minot (part 4) Cities
2025-03-28 13:41:49

yesterday:

  • three new "greenland" benchmarks: letter count, unit conversion, part-of-speech detection. ๐Ÿ’ก it should not be a surprise to the contemporary reader that the models struggle the most with "letter count" - how many "r"s are in strawberry.

still to do:

  • code cleanup ๐Ÿ’ก the "run" method is written in slightly different form seven times
  • more benchmarks
  • dashboard UI improvements
minot (part 3) Cities
2025-03-26 20:51:59

today: https://spaceship.computer/greenland/model_summary.html

These are "proficiency" metrics. ๐Ÿ”ฅ although, every time I use the word "proficiency" I want to change it

They are simple tasks, currently: translate a word, choose a definition, choose an antonym, find the misspelled word. And, for the >4B models, as long as the model knows the language, it does fairly well. The 1B models do have some difficulties.

The timing data is interesting. It is, roughly, a linear relation to model size. The 9B models are about 4 times slower than the 1B models. Phi-4 (the largest model tested) is also very clearly the slowest model.

Some of the models I was looking at before (Granite, ExaONE, Hermes, Tulu, Mistral) did not make this round of tests. For Mistral, the 12B model is too old, and their newest release, at 24B, is too large. The others didn't distinguish themselves enough from similar Llama models to be worth my time (and hard-drive space).

remaining todo:

  • standardize the logging of prompts and responses. the full text โš™๏ธ that is, including the system prompt should be stored.
  • fix the benchmarks. some of the definitions are too similar. โš™๏ธ previously we had kingdom and realm as choices. now the closest is honest and sincere. some of the translations are still a bit rough. ๐Ÿ’ก the translation of "beautiful" into French is beau/belle, the LLMs are very reasonably just returning "beau" as the translation
  • fix the model warming. Just calling the "warm model" function correctly doesn't do enough warming.
  • add additional tests. hopefully now it will take less than 1 hour to make new tests.

some of the suggestions regarding new tests:

Part of Speech Tagging - Present a sentence and ask the model to identify the part of speech (noun, verb, adjective, etc.) for a specific word.

Unit Conversion - Test ability to convert between simple units (kilometers to miles, pounds to kilograms).

Analogies - Simple analogies like "day is to night as hot is to ___".

Tense Transformation - Provide a sentence in one tense and ask the model to convert it to another tense.

Active/Passive Voice Conversion - Convert sentences between active and passive voice.

minot (part 2) Cities
2025-03-25 17:42:51

for today:

  • one investment meeting โš™๏ธ as always, further information withheld
  • tutoring ๐Ÿ’ก details withheld, but some information may make it to a post ๐ŸŒŽ due to staff illness, tutoring was canceled
  • maybe try again to get Claude to see the light about the lexer/parser system ๐Ÿ’ก currently, the lexer is looking ahead to capture an entire text block for the emphasis tag, rather than letting the parser do this
  • add HTML metadata linking to the RSS feed here.
  • maybe get the LLM dashboard available publicly ๐Ÿ’ก but first, I would like 3 or 4 more of the "proficiency" tests on the dashboard.

----

apparently https://github.com/ollama/ollama/issues/7978 is fixed. This was the "Ollama always returns structured-JSON responses in alphabetical order" bug. Which, when the two fields were thought and answer, was a major problem.

I need Claude to remove the code that was added to work-around this problem. Which is at least a 1-hour time commitment.

At that point, Claude should be able to construct new "proficiency" benchmarks fairly easily. ๐Ÿ’ก with a perfect system, it would take only 5 minutes per test. In the current world, I am hoping for 20-30 minutes.

Show Newer Messages Show Older Messages