David Rozado

I, for one, would like to see an open-source model similar to Stable Diffusion, but for text. It would be a great way to empower general folk without having to pay OpenAI, and not have to worry about the LLM's belief system, which is conservative-biased in the case of ChatGPT[1] (HN discussion[2]). [1] https://davidrozado.substack.com/p/openaicms [2] https://news.ycombinator.com/item?id=34625001

>Are you tracking actual identity politics or the term "identity politics"? The term "identity politics" is not being tracked. Rather, terminology used by modern left-wing idpol ideology is being tracked. See here: https://davidrozado.substack.com/p/new-york-times-word-usage-frequency >the meaning of the term applies just as much to ending slavery, womens' suffrage and civil rights movements. I'm not sure that's true. E.g. Martin Luther King Jr spoke of the "magnificent words" in the Constitution and the Declaration of Independence. Modern activists would say: "Written by dead pale male slaveholders. We need more diversity! Where are the voices of women and POC?" For King, ideas took precedence over identities. For modern activists, it's the opposite. The American Anti-Slavery Society was predominantly white. That's puzzling for a movement driven by identity politics. It does make sense for a movement driven by universal humanitarian ideals. In any case, if you still think I'm wrong, and identity politics is an essential force behind trends such as the civil rights movement -- then I suppose you'll be happy that it's being adopted by the political right in the United States? Since it's got such a great historical track record, surely results will be good? ;-) >Otherwise, you might as well argue that fake news only existed from 2016 onwards, because that's when Google Trends says it did. This is a bad analogy, because fake news itself doesn't use the term…

Here is an investigation of how different queries are classified as hateful vs not hateful in ChatGPT: https://davidrozado.substack.com/p/openaicms

Given the mountain of evidence for Wikipedia's left wing bias[1][2][3][4], it seems something ought to change, but hard to answer "how?" I expect to receive many downvotes for this, despite providing high-quality and authoritative sources. Which is not unlike what happens on Wikipedia :) [1]: https://davidrozado.substack.com/p/is-wikipedia-politically-biased [2]: https://www.piratewires.com/p/how-the-regime-captured-wikipedia [3]: https://larrysanger.org/2020/05/wikipedia-is-badly-biased/ [4]: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5022797

A quantitative look which might interest you: https://davidrozado.substack.com/p/is-wikipedia-politically-biased

ChatGPT (at least earlier versions) was deliberately fine-tuned to strongly favor to one side of the American culture war. See this paper on the its political bias: https://www.mdpi.com/2076-0760/12/3/148. Also, see David Rozado's Substack for an attempt at creating a chatbot with the opposite biases: https://davidrozado.substack.com/p/rightwinggpt.

I think the only thing this article shows it the authors ability to use chatGPT to draw views as a subject. For example, just a few weeks ago he wrote an article describing how ChatGPT has been updated so that it doesn’t show left wing bias [o] https://davidrozado.substack.com/p/chatgpt

No. Unless the apolitical version is a paid option. https://davidrozado.substack.com/p/openaicms

Yes, but AI was supposed to improve things, not make the gap wider https://davidrozado.substack.com/p/openaicms

A recent update showed it's moving more towards the center, which gives me some hope that they're trying to make it more objective: https://davidrozado.substack.com/p/chatgpt

Same author did an analysis [1] on where ChatGPT falls on the political spectrum and, correlating closely with the results listed here, it is lib-left. [1] https://davidrozado.substack.com/p/political-bias-chatgpt

Good catch. Article here: https://davidrozado.substack.com/p/chatgpt

Link to blog post: https://davidrozado.substack.com/p/chatgpt

I put together a small interactive tool on Hugging Face Spaces that lets you compare the political compass of four models (OpenAI GPT-4o, DeepSeek DeepSeek-chat-v3-0324, X-ai Grok-beta, MistralAI Mistral-large-2411) on a set of 62 political questions. The interface scores models along economic and social axes using political quiz data (Libertarian/Authoritarian, Left/Right), aiming to reproduce works like https://davidrozado.substack.com/p/new-results-of-state-of-the-art-llm… or https://trackingai.org/political-test. The purpose of the interface is more pedagogical (I do trainings on AI for the public at large) than political, and to offer a way to raise awareness about political biases between models or languages for non expert crowds. It is still interesting though to observe how current state-of-the-art LLMs often lean left-of-center, and that some models (notably OpenAI's and DeepSeek's) seem to display _stronger left polarization_ when prompted in *French* as opposed to English. The project is open-source on GitHub, and it should be easy to adapt it to other languages or even other ideological scales. Contributions or feedback welcome! https://github.com/Yannael/ai-political-bias

A genuine answer, how about looking up some studies on this subject? Not those done by Wikipedia of course, they claim to be politically neutral after all. Here's a few, from https://www.allsides.com/blog/wikipedia-biased Six studies, including two from Harvard researchers, have found a left-wing bias at Wikipedia: A 2024 analysis [1] by researcher David Rozado that used AllSides Media Bias Ratings [2] found Wikipedia associates right-of-center public figures with more negative sentiment than left-wing figures, and tends to associate left-leaning news organizations with more positive sentiment than right-leaning ones. A Harvard study [3] found Wikipedia articles are more left-wing than Encyclopedia Britannica. Another paper [4] from the same Harvard researchers found left-wing editors are more active and partisan on the site. A 2018 analysis [5] found top-cited news outlets on Wikipedia are mainly left-wing. Another analysis [6] using AllSides Media Bias Ratings found that pages on American politicians cite mostly left-wing news outlets. American academics found [7] conservative editors are 6 times more likely to be sanctioned in Wikipedia policy enforcement. There are far more sources out there. If I show examples of biased pages - the one on Antifa is a good example - this will just devolve into a quibble about this or that sentence. [1]…

There's lots of data, yes. Rozado has many n-gram plots that show this happening around 2010 for some terms and 2012-ish for other terms: https://davidrozado.substack.com/p/ppdwnmd The huge spikes in terms like inclusion, racist, sexist, climate change, global warming, etc is clear, all starting around either 2010 or 2012 (except for global warming, which spiked a few years earlier and then was replaced almost overnight in 2010 by the vaguer term climate change, which shows massive growth in the last 10 years). A clear trend of increasing prevalence of prejudice related terms is apparent with words such as racist or sexist increasing in usage between 2010 and 2019 by 638% and 403% in The New York Times or 514% and 141% respectively in The Washington Post. >> "should I just listen to a fat white man screaming on talk radio" I thought that would be considered racist fat shaming these days, but I guess it doesn't count if the person you're offending isn't on your team?

I think the Substack article is a better venue for the audience here: https://davidrozado.substack.com/p/political-bias-chatgpt I appreciate the work done here. We definitely need to understand more about the bias in such tools and as the other comment here noted, the zeitgeist of modern internet is left-leaning so it is unsurprising that this trend is also found in ChatGPT. However, I would say that these online tests are not necessarily robust. Saying 14 out of 15 tests went a certain way when you include low-quality efforts such as this one (https://www.proprofs.com/quiz-school/story.php?title=what-is-your-poli… is unhelpful. Another issue is that this is very American-centric. The average center-right European would likely fall in the left spectrum in the US. Considering training data is worldwide, your results might simply reflect the fact that the US is a right-leaning country by Western standards.