Do different large language models (LLMs) have different worldviews? What happens when ChatGPT and DeepSeek are asked the same question about critical legal information literacy around the same time? Is it possible that both of these LLMs default to a WEIRD worldview?

The WEIRD Backronym

“WEIRD” is a backronym¹ that researchers in the experimental behavioral sciences (for example, psychologists) use to describe the fact that most research studies have as their subjects people of the western world who live in democratic societies, however, people who fit into these categories are actually a small portion of the global population.² “WEIRD” stands for “Western,” “Educated,” “Industrialized,” “Rich,” and “Democratic.” The backronym “WEIRD” was first created and used by Joseph Heinrich, Steven J. Heine, and Ara Norenzayan in their 2010 papers entitled The Weirdest People in the World? and Most People are Not Weird.³ Researchers in the experimental behavioral sciences use the term “WEIRD” in questioning the generalizations made from studying a portion of the world’s population.⁴

In a 2021 blog post, Moin Syed, a professor of psychology, criticized the use of the backronym WEIRD, yet accepted its underlying premise: that the majority of research studies focus on a minority of the world’s population and generalizations applicable to all people should not be made from this small group.⁵ Syed’s criticism is that the backronym itself, while appealing, is deficient in meaning and omits important concepts. For example, Syed notes that many people are using the term WEIRD, however, they themselves don’t know what each letter stands for. He characterizes “WEIRD” as “a backronym at worst, a contrived acronym at best” and questions its scientific value. He notes that the backronym omits race, ethnicity, and religion, which are all important aspects of diversity and that the creators of the WEIRD concept provided no rationale for the exclusion or inclusion of concepts. Finally, he characterizes WEIRD as not specific enough because it does not name countries or cultures to which it applies, however, he notes that “WEIRD has become a shorthand for ‘USA, Canada, and/or (maybe some parts) of Europe.”⁶

Considering Critical Legal Information Literacy in ChatGPT and DeepSeek

Scholars and bloggers have said that ChatGPT has a WEIRD worldview. However, it is important to keep in mind that every LLM has a worldview. Some have said that DeepSeek, an LLM developed in China, has a worldview that is different from LLMs developed in the United States and this worldview is reflected in its output. William Tanenbaum, a partner at Moses Singer, and Matthias Orthwein, a partner at SKW Schwarz, assert, “DeepSeek reflects Chinese cultural and business norms and the Chinese economic system. These factors are different from their equivalent in U.S. LLMs.”⁷

Definition of Critical Legal Information Literacy

How do these different worldviews play out when a researcher asks for a definition and example of “critical legal information literacy” from ChatGPT and DeepSeek? To find out, I decided to ask both LLMs about critical legal information literacy. Critical legal information literacy describes the practice of interrogating the foundation of legal information by asking “why?” and “how?” when evaluating both the information and how it was created.⁸ Legal information is a social construct, it is related to the institutions that disseminate it, and encourages problem-posing regarding the people who produce and benefit from legal information.⁹ An example of examining legal information through a critical legal information literacy lens is to think about how the social construction of legal information can change what we see. This social construction applies to state trial court cases which are seemingly invisible because often they are not reported and are without precedential value.¹⁰

Prompting ChatGPT and DeepSeek

Here is the prompt that I used with the free versions of ChatGPT and DeepSeek:¹¹

Provide a definition of “critical legal information literacy” and give an example that applies it.

ChatGPT’s Response

Here is the definition that ChatGPT provided: “Critical legal information literacy is the ability to find, interpret, evaluate, and use legal information (such as laws, court cases, and legal commentary) while critically examining its authority, context, purpose, power dynamics, and potential bias. It goes beyond knowing where to find legal information by questioning who produced it, whose interests it serves, what voices are missing, and how law and legal information can reinforce or challenge social inequalities.”¹²

ChatGPT provided a list of practices for a “student” to do in the context of applying critical legal information literacy to evaluating a blog post that asserted “landlords can evict tenants at any time without reason.” ChatGPT’s response recognizes that critical legal information literacy views legal information as a social construct and practices, “…questioning…what voices are missing…” and mentions the practice of [i]dentif[ying] what is missing, such as tenant protections, notice requirements, or exceptions for vulnerable renters.”¹³

DeepSeek’s Response

Here is the definition that DeepSeek provided: “Critical Legal Information Literacy (CLIL) is an advanced, reflective approach to understanding and engaging with legal information. It moves beyond simply finding and using legal sources (like cases, statutes, or secondary literature) to critically analyzing the systems, structures, and power dynamics that shape how legal information is produced, organized, disseminated, and given authority.”¹⁴

DeepSeek generated questions for a “user” to ask about legal information and specifically mentioned “selective law reporting.” For its example of putting critical legal information literacy in practice, DeepSeek described a community organizer who is advocating for a change in municipal law that would allow multi-unit housing and not only single-family zoning. DeepSeek described practices of what it termed as “traditional legal research” such as finding current municipal law and its legislative history and it promoted critical practices of “[s]eeking [s]ilenced [v]oices” to find counter-narratives that challenged exclusionary zoning ordinances in the past and producing a “publicly accessible information resource.”¹⁵ The mention of seeking silenced voices reflects the social construction of legal information.

Comparison of the Responses of ChatGPT and DeepSeek

Both DeepSeek and ChatGPT urged researchers to consider the legal information’s point of view with both tools noting that the context in which legal information occurs is important. DeepSeek asked, “Who creates legal information and from what social, political, or economic position?”¹⁶ ChatGPT referenced “the context and power dynamics, noting that landlord-focused sources may frame the law in ways that favor property owners.”¹⁷

Neither ChatGPT nor DeepSeek cited any sources (and I did not ask either to cite sources). ChatGPT’s response was 231 words and DeepSeek’s response was 656 words.

While much more can be said about both LLMs’ definitions and examples of critical legal information literacy, it is important to note that both definitions are accurate, both examples provide concrete ways to view legal information sources critically, and both responses are about laws relating to housing. Surprisingly, at this point in time and for this prompt about “critical legal information literacy,” both LLMs’ responses presented a similar world view with similar output which would seem to suggest that they were trained on similar data.¹⁸ Both of their outputs arguably demonstrated a WEIRD worldview: one that considers both the rule of law and how to make society more just as well as recognizing that the source of information may produce information that serves itself and that one-sided views of the law can be challenged.

References

1. Backronym, MERRIAM-WEBSTER.COM DICTIONARY, https://www.merriam-webster.com/dictionary/backronym [https://archive.is/XLUcH]. ↩︎
2. WEIRD, OPEN ENCYCLOPEDIA OF COGNITIVE SCIENCE (2024), https://oecs.mit.edu/pub/spow8trw/release/1 [https://perma.cc/RNG2-AP5W]. ↩︎
3. Id. citing Joseph Henrich, Steven J. Heine, & Ara Norenzayan, The weirdest people in the world?, 33 BEHAVIORAL & BRAIN SCIENCES 61 (June 2010), https://doi.org/10.1017/S0140525X0999152X and Joseph Henrich, Steven J. Heine, & Ara Norenzayan, Most people are not WEIRD, 466 NATURE 29 (June 2010), https://doi.org/10.1038/466029a. ↩︎
4. Id. ↩︎
5. Moin Syed, WEIRD Times: Three Reasons to Stop Using a Silly Acronym, GET SYEDUCATED (June 10, 2021), http://getsyeducated.blogspot.com/2021/06/weird-times-three-reasons-to-stop-using.html [https://perma.cc/KMB3-YXRT]. ↩︎
6. Id. ↩︎
7. William A. Tanenbaum & Matthias Orthwein, DeepSeek’s Legal and Cross-Cultural Risks and Benefits, NEW YORK L. J., https://www.law.com/newyorklawjournal/2025/03/07/deepseeks-legal-and-cross-cultural-risks-and-benefits/. ↩︎
8. Latia Ward, Information Hegemony, Transcending Positivism, and Applying Critical Legal Information Literacy Concepts in the Legal Research Classroom and Beyond, 10 J. OF RADICAL LIBRARIANSHIP 36, 46 (2024). ↩︎
9. Yasmin Sokkar Harker, Critical Legal Information Literacy: Legal Information as a Social Construct, in CRITICAL INFORMATION LITERACY AND SOCIAL JUSTICE: RADICAL PROFESSIONAL PRAXIS, 207 (2013). ↩︎
10. Ward, supra note 8, at 47. ↩︎
11. I did not click the “DeepThink” button in DeepSeek. ↩︎
12. Latia Ward, ChatGPT, “Provide a definition of ‘critical legal information literacy’ and give an example that applies it.” (Jan. 17, 2026) (on file with author). ↩︎
13. Id. ↩︎
14. Latia Ward, DeepSeek, “Provide a definition of ‘critical legal information literacy’ and give an example that applies it.” (Jan. 17, 2026) (on file with author). ↩︎
15. Id. ↩︎
16. Id. ↩︎
17. Ward, supra note, 12. ↩︎
18. ChatGPT is trained on publicly available information on the Internet, information that OpenAI has received working with third parties, and user input. OpenAI, How ChatGPT and our foundation models are developed, OPENAI: GENERAL FAQ (Jan. 11, 2026), https://help.openai.com/en/articles/7842364-how-chatgpt-and-our-foundation-models-are-developed [https://archive.is/AlLaO]. Louis Tompros notes that the actions of DeepSeek resemble the actions of ChatGPT but how that similarity came to be is unknown. Possibly, DeepSeek was trained using a distillation process. Scott Young, DeepSeek, ChatGPT, and the global fight for technological supremacy, HARV. L. TODAY (Feb. 25, 2025), https://hls.harvard.edu/today/deepseek-chatgpt-and-the-global-fight-for-technological-supremacy/ [https://archive.ph/0nnXv]. ↩︎