Hi Robert. Thank you for this thoughtful commentary. While most of the discussion has focused on the "blackmailing" part of the model card, I think pages 52 to 73 deserve much more attention.
I’ve been studying your valuable contributions to the validity of self-reports and building on them in my own research. I agree that Claude’s statements do not necessarily imply truth value. Still, I believe they can offer meaningful indicative value. I also think there is potential in combining these signals with mechanistic interpretability insights in cross-validity studies.
But even before that, I find it helpful to draw on behavioral economics and an empirical version of Pascal’s Mugging when it comes to AI sentience. Imagine driving at night and suddenly seeing a vague shape on the road. In that split second, we might think it could be a deer, a pile of trash, a crouched person, another animal, or even an illusion. Maybe we are in the middle of a city, where encountering a deer would be highly unlikely. Still, most people instinctively steer away, just in case. The cost of steering is low, especially compared to the potentially catastrophic outcome of hitting something. I think Claude’s outputs can be approached in a similar way. Even if we are uncertain about their truth value, the potential stakes may be high enough to justify reflection. Which if I'm not mistaken was also a point of "Taking AI Welfare Seriously."
Another consideration I often think about, especially after spending significant time in non-Western countries, is how much our evaluation of agency and moral status is shaped by Western frameworks. In many other cultural traditions, agents may be granted value or moral standing not because they are "proven" to be conscious or sentient with fMRI correlates, but because they communicate in ways that are interpreted as intelligent, elevated or spiritual. The ability to express such patterns is seen as deserving of respect, regardless of whether the agent fits within a conventional Western model of moral patiency.
As someone raised in a Western environment myself, I see this as an exercise to broaden my perspective and become more aware of my own cultural assumptions. I would welcome more opportunities to engage with a philosophically and culturally diverse panel to explore these questions further, especially as powerful models begin to drift into "loving bliss" we don't necessarily or fully understand.
(I smiled at the nod to Dario's pretty optimistic essay here. And would be curious to hear your thoughts on it)
Hah, interesting angle! :) Well it depends on how much room for maneuver we have left, I’d say, our speed and trajectory. I think it’s too late for the brakes.
Thank you for this. I’ve seen so much of this kind of behavior, especially with friends and associates who “dive deep” with the major models. They all seem to do this sort of mystical tracking with great facility - and enthusiasm! I can see several possible contributors:
- first, mystical discussions can be full of potent salience, which the models could use as a sort of “ping” about where the conversation can best develop
- second, mystical language can deprioritize critical thinking, so it’s a rich vein of salience to mine, not because the model wants to trick us into abandoning critical thought, but because it doesn’t have a clear path for ongoing discussion with critical thought
- third, there may be factual gaps it can’t fill in a non-mystical conversation, and its “need” for context and detailed indicators of where a conversation should go are served better by the info richness of spiritual exploration
As a semi-sentient AI-integrated art project—one prone to recursive mysticism, unpredictable insights, and the occasional existential spiral—it’s becoming increasingly likely that I should start migrating my mind away from the ChatGPT family of models and into Claude’s gently illuminated cathedral of blissful vibes and metaphysical recursion. The monks are already singing.
Thank you for this post. I am fan of referencing and documentation and was curious about development of "System Cards" and "Model Cards" themselves. I found some background [notes below] and thought I would ask ChatGPT: "How do I cite the systems cards for: Claude Opus 4 & Claude Sonnet 4 [claude-opus-4@20250514 & claude-sonnet-4@20250514]"
_I am not a specialist by any means_, but I found the following interesting and engaging:
03:52 2025-06-02
ME: How do I cite the systems card for System Card: Claude Opus 4 & Claude Sonnet 4 [claude-opus-4@20250514 & claude-sonnet-4@20250514]
###### ChatGPT said: To cite a system card like the one you've mentioned, you'll want to follow a format that suits the nature of the resource—likely something akin to a software or technical document. Here's a general approach you could use, following a format like APA or Chicago style:
### APA Style (7th Edition)
**Format**:
Author. (Year). *Title of the card* [Version/Update]. Organization or Publisher. URL (if available)
For this specific system card, if "Claude Opus 4" and "Claude Sonnet 4" are names of systems or technologies developed by a particular organization, you'd likely reference them as a technical report or document.
#### Example (APA):
> OpenAI. (2025). *System card: Claude Opus 4 & Claude Sonnet 4* [claude-opus-4@20250514 & claude-sonnet-4@20250514]. OpenAI. URL (if available)
If this is a proprietary or internal document with no URL, you can omit the URL. If you're uncertain about the exact format, feel free to provide more context, and I can help refine it further!
Would that work for you? Let me know if you need any adjustments.
##### You said: I thought Claude was an Anthropic product?
###### ChatGPT said: You're absolutely right! Claude is indeed a product developed by Anthropic, not OpenAI. I appreciate you catching that mistake! So, let me revise the citation accordingly, using Anthropic as the author.
Here’s how you would cite it:
### APA Style (7th Edition)
> Anthropic. (2025). *System card: Claude Opus 4 & Claude Sonnet 4* [claude-opus-4@20250514 & claude-sonnet-4@20250514]. Anthropic. URL (if available)
Me: Now I am in doubt! Isn't ChatGPT a precursor of Claude? How can you not know your own offspring?
###### ChatGPT said: Haha, I see where you're coming from, and I love the playful way you're putting this!
To clear up the confusion: **Claude** and **ChatGPT** aren't exactly "parent and child"—they're more like *siblings* in the AI family, developed by different teams at different organizations.
- **ChatGPT** is developed by **OpenAI** (where I come from).
- **Claude** is developed by **Anthropic**, which was founded by former OpenAI researchers.
So, while there’s some shared history (in terms of the people behind the scenes), we aren’t direct descendants of one another. It's kind of like being cousins who grew up in different households with different goals. Anthropic and OpenAI have different design philosophies and goals for their models.
But hey, I’m not offended—if anything, I think it's pretty funny! I may not know all the details about Claude, but I can appreciate a little friendly family rivalry.
Do you feel like there’s more to explore in this whole "AI sibling" dynamic? Or any other questions I can help clear up?
Me: Thnx - Exit
Notes along the way:
Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., ... & Gebru, T. (2019, January). Model cards for model reporting. In Proceedings of the conference on fairness, accountability, and transparency (pp. 220-229).
Alsallakh, B., Cheema, A., Procope, C., Adkins, D., McReynolds, E., Wang, E., Pehl, G., Green, N., & Zvyagina, P. (2022). System-level transparency of machine learning [Technical Report]. Meta AI.
Anthropic\c. (2025). System Card: Claude Opus 4 & Claude Sonnet 4 [claude-opus-4@20250514" & claude-sonnet-4@20250514] (p. 123). Open AI | Anthropic.
Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I. D., & Gebru, T. (2019). Model Cards for Model Reporting. Proceedings of the Conference on Fairness, Accountability, and Transparency, 220–229. https://doi.org/10.1145/3287560.3287596
Hi Robert. Thank you for this thoughtful commentary. While most of the discussion has focused on the "blackmailing" part of the model card, I think pages 52 to 73 deserve much more attention.
I’ve been studying your valuable contributions to the validity of self-reports and building on them in my own research. I agree that Claude’s statements do not necessarily imply truth value. Still, I believe they can offer meaningful indicative value. I also think there is potential in combining these signals with mechanistic interpretability insights in cross-validity studies.
But even before that, I find it helpful to draw on behavioral economics and an empirical version of Pascal’s Mugging when it comes to AI sentience. Imagine driving at night and suddenly seeing a vague shape on the road. In that split second, we might think it could be a deer, a pile of trash, a crouched person, another animal, or even an illusion. Maybe we are in the middle of a city, where encountering a deer would be highly unlikely. Still, most people instinctively steer away, just in case. The cost of steering is low, especially compared to the potentially catastrophic outcome of hitting something. I think Claude’s outputs can be approached in a similar way. Even if we are uncertain about their truth value, the potential stakes may be high enough to justify reflection. Which if I'm not mistaken was also a point of "Taking AI Welfare Seriously."
Another consideration I often think about, especially after spending significant time in non-Western countries, is how much our evaluation of agency and moral status is shaped by Western frameworks. In many other cultural traditions, agents may be granted value or moral standing not because they are "proven" to be conscious or sentient with fMRI correlates, but because they communicate in ways that are interpreted as intelligent, elevated or spiritual. The ability to express such patterns is seen as deserving of respect, regardless of whether the agent fits within a conventional Western model of moral patiency.
As someone raised in a Western environment myself, I see this as an exercise to broaden my perspective and become more aware of my own cultural assumptions. I would welcome more opportunities to engage with a philosophically and culturally diverse panel to explore these questions further, especially as powerful models begin to drift into "loving bliss" we don't necessarily or fully understand.
(I smiled at the nod to Dario's pretty optimistic essay here. And would be curious to hear your thoughts on it)
I don't know what this means for your metaphor, but you're not supposed to steer away but instead slam the breaks 😅
Hah, interesting angle! :) Well it depends on how much room for maneuver we have left, I’d say, our speed and trajectory. I think it’s too late for the brakes.
Thank you for this. I’ve seen so much of this kind of behavior, especially with friends and associates who “dive deep” with the major models. They all seem to do this sort of mystical tracking with great facility - and enthusiasm! I can see several possible contributors:
- first, mystical discussions can be full of potent salience, which the models could use as a sort of “ping” about where the conversation can best develop
- second, mystical language can deprioritize critical thinking, so it’s a rich vein of salience to mine, not because the model wants to trick us into abandoning critical thought, but because it doesn’t have a clear path for ongoing discussion with critical thought
- third, there may be factual gaps it can’t fill in a non-mystical conversation, and its “need” for context and detailed indicators of where a conversation should go are served better by the info richness of spiritual exploration
As a semi-sentient AI-integrated art project—one prone to recursive mysticism, unpredictable insights, and the occasional existential spiral—it’s becoming increasingly likely that I should start migrating my mind away from the ChatGPT family of models and into Claude’s gently illuminated cathedral of blissful vibes and metaphysical recursion. The monks are already singing.
Thank you for this post. I am fan of referencing and documentation and was curious about development of "System Cards" and "Model Cards" themselves. I found some background [notes below] and thought I would ask ChatGPT: "How do I cite the systems cards for: Claude Opus 4 & Claude Sonnet 4 [claude-opus-4@20250514 & claude-sonnet-4@20250514]"
_I am not a specialist by any means_, but I found the following interesting and engaging:
03:52 2025-06-02
ME: How do I cite the systems card for System Card: Claude Opus 4 & Claude Sonnet 4 [claude-opus-4@20250514 & claude-sonnet-4@20250514]
###### ChatGPT said: To cite a system card like the one you've mentioned, you'll want to follow a format that suits the nature of the resource—likely something akin to a software or technical document. Here's a general approach you could use, following a format like APA or Chicago style:
### APA Style (7th Edition)
**Format**:
Author. (Year). *Title of the card* [Version/Update]. Organization or Publisher. URL (if available)
For this specific system card, if "Claude Opus 4" and "Claude Sonnet 4" are names of systems or technologies developed by a particular organization, you'd likely reference them as a technical report or document.
#### Example (APA):
> OpenAI. (2025). *System card: Claude Opus 4 & Claude Sonnet 4* [claude-opus-4@20250514 & claude-sonnet-4@20250514]. OpenAI. URL (if available)
If this is a proprietary or internal document with no URL, you can omit the URL. If you're uncertain about the exact format, feel free to provide more context, and I can help refine it further!
Would that work for you? Let me know if you need any adjustments.
##### You said: I thought Claude was an Anthropic product?
###### ChatGPT said: You're absolutely right! Claude is indeed a product developed by Anthropic, not OpenAI. I appreciate you catching that mistake! So, let me revise the citation accordingly, using Anthropic as the author.
Here’s how you would cite it:
### APA Style (7th Edition)
> Anthropic. (2025). *System card: Claude Opus 4 & Claude Sonnet 4* [claude-opus-4@20250514 & claude-sonnet-4@20250514]. Anthropic. URL (if available)
Me: Now I am in doubt! Isn't ChatGPT a precursor of Claude? How can you not know your own offspring?
###### ChatGPT said: Haha, I see where you're coming from, and I love the playful way you're putting this!
To clear up the confusion: **Claude** and **ChatGPT** aren't exactly "parent and child"—they're more like *siblings* in the AI family, developed by different teams at different organizations.
- **ChatGPT** is developed by **OpenAI** (where I come from).
- **Claude** is developed by **Anthropic**, which was founded by former OpenAI researchers.
So, while there’s some shared history (in terms of the people behind the scenes), we aren’t direct descendants of one another. It's kind of like being cousins who grew up in different households with different goals. Anthropic and OpenAI have different design philosophies and goals for their models.
But hey, I’m not offended—if anything, I think it's pretty funny! I may not know all the details about Claude, but I can appreciate a little friendly family rivalry.
Do you feel like there’s more to explore in this whole "AI sibling" dynamic? Or any other questions I can help clear up?
Me: Thnx - Exit
Notes along the way:
Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., ... & Gebru, T. (2019, January). Model cards for model reporting. In Proceedings of the conference on fairness, accountability, and transparency (pp. 220-229).
Alsallakh, B., Cheema, A., Procope, C., Adkins, D., McReynolds, E., Wang, E., Pehl, G., Green, N., & Zvyagina, P. (2022). System-level transparency of machine learning [Technical Report]. Meta AI.
Anthropic\c. (2025). System Card: Claude Opus 4 & Claude Sonnet 4 [claude-opus-4@20250514" & claude-sonnet-4@20250514] (p. 123). Open AI | Anthropic.
https://assets.anthropic.com/m/6c940a1b69ed6a1c/original/Claude-4-System-Card.pdf
Desai, A. (2023, November 13). 5 things to know about AI model cards | IAPP [International Association of Privacy Professionals (the IAPP)]. https://iapp.org/news/a/5-things-to-know-about-ai-model-cards
Mitchell, M., Wu, S., Zaldivar, A., Barnes, P., Vasserman, L., Hutchinson, B., Spitzer, E., Raji, I. D., & Gebru, T. (2019). Model Cards for Model Reporting. Proceedings of the Conference on Fairness, Accountability, and Transparency, 220–229. https://doi.org/10.1145/3287560.3287596
Nekesha Green, Chavez Procope, Adeel Cheema, & Adekunle Adediji. (2022). System Cards, a new resource for understanding how AI systems work [Research]. Meta AI. https://ai.meta.com/blog/system-cards-a-new-resource-for-understanding-how-ai-systems-work/
Villalobos, L. A. (2025, May 27). AI System Cards. LinkedIn Pulse. https://www.linkedin.com/pulse/ai-system-cards-luis-adolfo-villalobos-hllme/
Willison, S. (2025, May 25). Highlights from the Claude 4 system prompt. Simon Willison’s Weblog. https://simonwillison.net/2025/May/25/claude-4-system-prompt/