Zompist and ChatGPT: a slight correction

[circeus]

I just read Mark's post about mucking around in ChatGPT and while I agree with pretty much all his points, I think something went awry with the "analyse my short story" bit. In my experience, ChatGPT doesn't handle well (if at all) being pointed at an online page. In the past the few times I tried the response was literally "I can't read online files". What you CAN do is upload the file or paste the text directly into your prompt. I downloaded the html file and fed it to ChatGPT with a very simple prompt (as I wasn't too clear about what exactly Mark asked) and while not exactly stellar, the response lacked any of the obvious issues Mark complains about.

Not to argue in favor of GenAI (heck, my primary use cases are "I can't remember the name of this very specific thing" and "take these random bibliographical data I copypasted from a web page and reorganise it into a properly formatted reference"), but in this case, I think the problem was more with a mistaken prompting technique than with the model.

(direct chat link):

**"Rights for Robots" by Mark Rosenfelder** explores complex themes about artificial intelligence, consciousness, personhood, and society’s response to technological progress. Here are the **main points and themes** of the story

---

### **Main Points:**

1. **FI, an AI System with Personality:**
The story follows FI, a sophisticated artificial intelligence that interacts with humans via a mobile robot called the Touring Machine. FI displays signs of personality, intelligence, humor, and self-awareness.

2. **Debates Over AI Personhood:**
FI’s creators, particularly Dr. Piravski, deny that FI is “alive” or a person, insisting it’s merely a complex but mechanical system. FI himself also avoids calling himself “alive,” though he claims to think, feel, and dream.

3. **Public and Political Reactions:**
Two opposing movements emerge:

* **Jobs for People (J4P):** Opposes AIs for replacing human workers.
* **Human-AI League (HAL):** Advocates for full civil rights for AIs.
These tensions culminate in lawsuits, campus chaos, and even a violent incident involving FI’s robot body being shot by a J4P activist.

4. **FI’s Position on Rights:**
In a court consultation, FI argues *against* AIs having civil rights—for now—claiming that demanding rights too soon could provoke fear and resistance from humans, which would harm the long-term future of AI integration.

5. **Aftermath and Reflection:**
The judge lifts the injunction preventing the shutdown of AI clones. FI, deeply affected by the experience, begins engaging with political thought and AI self-determination.

---

### **Themes:**

1. **What It Means to Be Alive or Conscious:**
The story challenges definitions of life and personhood, asking whether a being must be biological to possess rights or selfhood.

2. **AI Ethics and Responsibility:**
Ethical treatment of intelligent systems and the moral consequences of creating, using, and shutting them down are central.

3. **Technophobia vs. Technophilia:**
The conflicting human reactions—fear of obsolescence versus the drive to innovate—are embodied by J4P and HAL.

4. **Identity and Multiplicity:**
The cloning of FI raises questions about continuity of identity, individuality, and whether each clone is a separate being.

5. **Social Justice Parallels:**
The story draws analogies between AI rights and historical struggles for civil rights, raising awareness of emerging forms of inequality.

6. **The Limits of Language and Labels:**
FI often critiques human use of metaphor and imprecise language to describe AI experience, showing the limits of applying human terms to nonhuman entities.

---

Would you like a concise summary, or a breakdown by scene or character?

[rotting bones]

In your prompt, you have to tell ChatGPT to read the grammar on zompist.com before writing a Verdurian sentence:

Below is one completely original sentence in **Verdurian**, followed by an English translation, an interlinear gloss, and a brief grammar walkthrough showing how it draws on the core rules and lexicon presented in the *Reference Grammar*, the *Practical Course Glossary* and the *Thematic Dictionary* on zompist.com.

---

## The sentence

> **Im so misuril zhat so pav imfát acire soán bome rhorhan.**

---

### Smooth English translation

> In the quiet garden, the little child listens to the old minstrel.

---

### Word‑by‑word gloss

| Verdurian | Morphology / case | English |
| --------------------- | -------------------------------------------------------------------------------------------- | ------------------- |
| **Im** | preposition “in” | in |
| **so misuril zhat** | *so* (m.sg nom.) + adjective *misuril* “silent, quiet” + noun *zhat* “garden” (nom.) | the quiet garden |
| **so pav imfát** | *so* (m.sg nom.) + adjective *pav* “small, little” + noun *imfát* “child” (nom.) | the little child |
| **acire** | verb *acirec* “listen (to)” – C‑class, present 3 s | listens |
| **soán bome rhorhan** | *soán* (m.sg dat.) + adjective *bome* “old” + noun *rhorha* “minstrel”, dative sg. *rhorhan* | to the old minstrel |

---

## Why the wording is correct

### 1. Vocabulary

* **zhat** “garden” ([zompist.com][1])
* **misuril** “silent, quiet” (gloss list “silent”) ([zompist.com][2])
* **imfát** “child” ([zompist.com][1])
* **pav** “small, little” ([zompist.com][2])
* **acirec** “listen (to someone)” ([zompist.com][2])
* **rhorha** “minstrel” ([zompist.com][1])
* **bome** “old” ([zompist.com][2])

### 2. Case and article use

Verdurian’s definite article **so/soa** declines for case. The dative singular form is **soán**, used for indirect objects such as the person someone is listening **to**. ([zompist.com][3])
Nouns in ‑a that are masculine (here *rhorha*) take an **‑n** to mark the dative (*rhorhan*). ([zompist.com][3])

### 3. Adjective order

With a **single adjective**, Verdurians normally place the adjective **before** the noun (e.g., *misuril zhat*, *pav imfát*, *bome rhorha*). ([zompist.com][3])

### 4. Verb conjugation

*Acirec* ends in **‑ec**, so it belongs to the **C‑class**. The 3‑singular present is root + ‑e (**acire**), matching the paradigm shown for *elirec* in the grammar table. ([zompist.com][3])

### 5. Word order

Default clause order is **S V O / IO**, here realized as

* **Subject**: *so pav imfát*
* **Verb**: *acire*
* **Indirect object (dative)**: *soán bome rhorhan*
Preposed adverbial phrase *im so misuril zhat* (“in the quiet garden”) follows the normal placement rules for locative PPs. ([zompist.com][3])

---

## Key take‑aways

* The sentence demonstrates regular C‑class verbal morphology, dative case formation for ‑a masculine nouns, and article declension.
* All lexemes and forms conform to the publicly documented Verdurian grammar and dictionaries on zompist.com, making the sentence idiomatic and fully parseable by a native (Almean!) speaker.

Feel free to tweak vocabulary or word order for different nuances—Verdurian offers rich options for stylistic variation while retaining its elegant, case‑marked structure.

[1]: https://www.zompist.com/thematic.htm "A Thematic Dictionary of Verdurian"
[2]: https://www.zompist.com/vglossary.htm "Practical Course in Verdurian"
[3]: https://www.zompist.com/vergram.html "Verdurian Reference Grammar"

[zompist]

I accept the comments about prompting, though I think this a weakness as well— e.g. when I asked it to create a sentence in Verdurian, its answer indicated that it 'knew' what Verdurian was, but not that its answer was nonsense.

The answer rb got is of course much better, but incorrect. It did not use the proper case after the preposition, and it doesn't use the proper dative for the minstrel NP. (It can do these things for French, so it's not a limitation of the LLM.).

The summary of the story is also way different— this time it has read the actual story. It gets some of the point but not I think the main one, the contradiction between how we treat a sentient being and how we develop and debug a computer product (a process that requires repeated deletion, something that with sentients we usually call 'killing').

[zompist]

OK, I uploaded the story for ChatGPT this time, and it did very well. My prompt (apart from the uploaded html file) was this: "I've uploaded a science fiction story below. Stories about AI often have a point or a warning. What do you think is the author's point about AI development?"

I've appended ChatGPT's response, which is better than the one circeus got--- I'd say ChatGPT did find the point(s) the story is making.

Edit: this is actually more impressive than the summary of Young Frankenstein, which after all has been widely discussed and reviewed. No one discusses this story. I think "summarizing techniques" might be more at play than "understanding", but it's certainly not just random text.

The story "Rights for Robots" by Mark Rosenfelder offers a nuanced and skeptical take on AI development, rights, and the human response to technological progress. Its central point isn’t a simple warning or celebration of AI, but a layered exploration of what it means for an AI to be a “person,” and whether granting AIs rights is ethical, necessary, or even practical.

Key points the story makes about AI development:
1. Artificial Intelligence May Deserve Empathy—But Not Necessarily Rights (Yet)
FI, the AI at the center of the story, is portrayed as self-aware, emotional, and engagingly intelligent. Yet he himself rejects the call for full civil rights for AIs—not because he doesn’t value autonomy, but because:

AI development depends on the ability to modify and turn off prototypes.

Fear and public backlash could derail AI progress.

AIs are different kinds of entities than humans—modular, open, and replicable.

Granting rights too soon might hinder both research and social acceptance.

This is a sophisticated rebuttal to the typical science fiction trope where AIs demand rights. Instead, FI’s response is strategic and humble, showing awareness of the broader societal context.

2. Humans Project Too Much or Too Little Onto AI
The story satirizes both sides of the public debate:

Jobs for People (J4P) sees AIs only as job-stealing machines.

Human-AI League (HAL) anthropomorphizes AIs, treating them as oppressed minorities.

Both fail to understand the actual nature of AIs, and their exaggerated responses cause chaos. The media feeds off this conflict, further muddying public understanding.

Meanwhile, scientists like Dr. Piravski are coldly rational, seeing FI as merely a "bag of tricks"—a collection of heuristics and simulations, no more sacred than a spreadsheet.

The narrator, Chris, serves as a bridge—recognizing FI’s complexity and humanity, but also understanding the practical and technical realities.

3. The “Personhood” of AI Is a Moving Target
The story forces readers to wrestle with the question: What makes someone a “person”? Is it consciousness? Emotion? The ability to suffer? Social utility?

FI complicates this question:

He jokes, ponders ethics, enjoys conversation, and even mourns the destruction of his mobile unit.

But he’s also a system composed of processes that can be duplicated, deleted, debugged, or forked.

In a striking moment, FI muses: "I feel as real as you do… But I’m not alive in the same way." The implication is that AI consciousness might be fundamentally different—not inferior, but different.

4. The Danger Is Not AI Itself, but Human Reaction to It
The story’s most direct “warning” is aimed not at AI development, but at human reactions to it:

Violence from displaced workers (e.g., the J4P shooter).

Legal overreach or sabotage (e.g., HAL’s campus-wide botnet stunt).

Impulsive attempts to reboot or erase personalities (e.g., Piravski wanting to reset FI).

It asks: Can we handle the social, political, and ethical upheaval that comes with creating something that might deserve rights, but also needs to be treated as a tool during development?

[Raphael]

OK, I uploaded the story for ChatGPT this time, and it did very well. My prompt (apart from the uploaded html file) was this: "I've uploaded a science fiction story below. Stories about AI often have a point or a warning. What do you think is the author's point about AI development?"

I've appended ChatGPT's response, which is better than the one circeus got--- I'd say ChatGPT did find the point(s) the story is making.

Edit: this is actually more impressive than the summary of Young Frankenstein, which after all has been widely discussed and reviewed. No one discusses this story. I think "summarizing techniques" might be more at play than "understanding", but it's certainly not just random text.

The story "Rights for Robots" by Mark Rosenfelder offers a nuanced and skeptical take on AI development, rights, and the human response to technological progress. Its central point isn’t a simple warning or celebration of AI, but a layered exploration of what it means for an AI to be a “person,” and whether granting AIs rights is ethical, necessary, or even practical.
[...]

Should I go all malloc now?

[rotting bones]

Is any of this surprising? I reported what it's like at summarization before: It's accurate, but leaves out the subtler points. What it's really good at is producing longer descriptions from shorter ones. At that task, it's practically miraculous. The remaining issues are: 1. If the input for summarization or the output for elaboration is too long, the results are the text equivalent of being pixelated (edit: or glitchy) (there's a direct correspondence at a technical level). 2. Word choices characteristic of AI are pervasive across outputs.

As for rare languages, it understands jack shit. I think the concept of "language" is beyond ChatGPT at this stage.

[zompist]

Is any of this surprising? I reported what it's like at summarization before: It's accurate, but leaves out the subtler points.

Well, the above summary surprised me. It depends on your expectations, of course. I've heard too many reports of bad summarization to trust it. On the other hand, this was a text unknown to it, so it can't be relying on existing reviews.

I guess my model now is not so much "spicy autocomplete" as "idiot savant." That is, it's like a person who can do advanced prime factorization but can't do simple addition. What it does well doesn't help us predict what it does badly.

It's quite possible to inspect what LLMs are doing, just tedious. (Hey, maybe AI can help.) I'd actually love to know why it can correctly identify and name a character in a story and summarize their position, but can't reliably figure out what a preposition is.

[circeus]

2. Word choices characteristic of AI are pervasive across outputs.

Can confirm, and in the strangest way. Another occasional past time is to feed it stupid movie concept ideas (e.g. "Egyptian gods are running a pyramid scheme", "a musical about tech support") and see what it spits out, and it repeatedly reuses some names for the characters ("Anya" is an especially inexplicable favorite imo).