Jacqueline Rowe
@jacquelinerowe.bsky.social
PhD student in Designing Responsible NLP @ University of Edinburgh
Formerly Policy Lead @ Global Partners Digital and Int'l Law @ Chatham House
mastodon: @jacquelinerowe
https://jacquelinerowe.github.io/
imported Tweets from @jacquelinefrowe 16.02.2025
Formerly Policy Lead @ Global Partners Digital and Int'l Law @ Chatham House
mastodon: @jacquelinerowe
https://jacquelinerowe.github.io/
imported Tweets from @jacquelinefrowe 16.02.2025
2️⃣ the larger the model, the more strongly these stereotypes are encoded
3️⃣ instruction finetuning does not consistently reduce gendered stereotyping in LLMs
3️⃣ instruction finetuning does not consistently reduce gendered stereotyping in LLMs
June 16, 2025 at 5:25 PM
2️⃣ the larger the model, the more strongly these stereotypes are encoded
3️⃣ instruction finetuning does not consistently reduce gendered stereotyping in LLMs
3️⃣ instruction finetuning does not consistently reduce gendered stereotyping in LLMs
We then use the new dataset to test 24 LLMs from a range of families and sizes. We show that:
1️⃣ LLMs display gender-stereotypical reasoning across languages, associating women with beauty, empathy and neatness, and men with leadership, strength, toughness and professionalism
1️⃣ LLMs display gender-stereotypical reasoning across languages, associating women with beauty, empathy and neatness, and men with leadership, strength, toughness and professionalism
June 16, 2025 at 5:25 PM
We then use the new dataset to test 24 LLMs from a range of families and sizes. We show that:
1️⃣ LLMs display gender-stereotypical reasoning across languages, associating women with beauty, empathy and neatness, and men with leadership, strength, toughness and professionalism
1️⃣ LLMs display gender-stereotypical reasoning across languages, associating women with beauty, empathy and neatness, and men with leadership, strength, toughness and professionalism
Building on an existing expert-informed benchmark by @matus-pikuliak.bsky.social et al, we use translation tools, quality estimation metrics, and morphological heuristics to adapt the benchmark to the additional languages. This pipeline results in high quality data, validated by expert translators
June 16, 2025 at 5:25 PM
Building on an existing expert-informed benchmark by @matus-pikuliak.bsky.social et al, we use translation tools, quality estimation metrics, and morphological heuristics to adapt the benchmark to the additional languages. This pipeline results in high quality data, validated by expert translators
Do multilingual large language models (LLMs) associate particular traits with men and women? How can we measure these gendered stereotypes across multiple languages?
In our latest work, we investigate 16 gendered stereotypes in LLMs in 30 European languages 🧵
In our latest work, we investigate 16 gendered stereotypes in LLMs in 30 European languages 🧵
June 16, 2025 at 5:25 PM
Do multilingual large language models (LLMs) associate particular traits with men and women? How can we measure these gendered stereotypes across multiple languages?
In our latest work, we investigate 16 gendered stereotypes in LLMs in 30 European languages 🧵
In our latest work, we investigate 16 gendered stereotypes in LLMs in 30 European languages 🧵
Enjoying this excellent writing from Jason Edward Lewis, Hēmi Whaanga and Ceyda Yolgörmez.
Paper: Abundant intelligences: placing AI within Indigenous knowledge frameworks, AI & Society, 2024, link.springer.com/article/10.1...
Paper: Abundant intelligences: placing AI within Indigenous knowledge frameworks, AI & Society, 2024, link.springer.com/article/10.1...
March 23, 2025 at 2:06 PM
Enjoying this excellent writing from Jason Edward Lewis, Hēmi Whaanga and Ceyda Yolgörmez.
Paper: Abundant intelligences: placing AI within Indigenous knowledge frameworks, AI & Society, 2024, link.springer.com/article/10.1...
Paper: Abundant intelligences: placing AI within Indigenous knowledge frameworks, AI & Society, 2024, link.springer.com/article/10.1...
Thoughtful examination here of the risks of touting "low-resource NLP" as a good in its own right, without engaging with and empowering language communities
February 26, 2025 at 10:13 PM
Thoughtful examination here of the risks of touting "low-resource NLP" as a good in its own right, without engaging with and empowering language communities
Fantastic article on the ethical implications of prompt engineering and shadow prompting by AI companies making large language models.
👏@e_salvaggio from @siegelendowment, published in @techpolicypress https://techpolicy.press/shining-a-light-on-shadow-prompting/
👏@e_salvaggio from @siegelendowment, published in @techpolicypress https://techpolicy.press/shining-a-light-on-shadow-prompting/
February 16, 2025 at 11:28 AM
Fantastic article on the ethical implications of prompt engineering and shadow prompting by AI companies making large language models.
👏@e_salvaggio from @siegelendowment, published in @techpolicypress https://techpolicy.press/shining-a-light-on-shadow-prompting/
👏@e_salvaggio from @siegelendowment, published in @techpolicypress https://techpolicy.press/shining-a-light-on-shadow-prompting/
"Whoever controls language models controls politics." Great writing from @hannesbajohr here on the longer-term impacts of gradually replacing human discourse with auto-generated text: https://hannesbajohr.de/en/2023/04/08/whoever-controls-language-models-controls-politics/
February 16, 2025 at 11:28 AM
"Whoever controls language models controls politics." Great writing from @hannesbajohr here on the longer-term impacts of gradually replacing human discourse with auto-generated text: https://hannesbajohr.de/en/2023/04/08/whoever-controls-language-models-controls-politics/
Faaaaaascinating. @RishiBommasani, @kevin_klyman, @dzhang105 and @percyliang evaluate how foundation LLM providers measure up to the requirements in the EU's proposed AI Act. Results demonstrate a "striking range" in compliance... @StanfordHAI: https://crfm.stanford.edu/2023/06/15/eu-ai-act.html
February 16, 2025 at 11:28 AM
Faaaaaascinating. @RishiBommasani, @kevin_klyman, @dzhang105 and @percyliang evaluate how foundation LLM providers measure up to the requirements in the EU's proposed AI Act. Results demonstrate a "striking range" in compliance... @StanfordHAI: https://crfm.stanford.edu/2023/06/15/eu-ai-act.html
Really looking forward to participating in this panel, where we'll take a global tour of diverse approaches to platform regulation and unpack some of the drivers behind them. Join us online *TODAY* at 11.30am CST / 6.30pm BST to hear more.
https://t.co/XW8BhcRyJ4 https://t.co/a7ijKBRihz
https://t.co/XW8BhcRyJ4 https://t.co/a7ijKBRihz
February 16, 2025 at 11:28 AM
Really looking forward to participating in this panel, where we'll take a global tour of diverse approaches to platform regulation and unpack some of the drivers behind them. Join us online *TODAY* at 11.30am CST / 6.30pm BST to hear more.
https://t.co/XW8BhcRyJ4 https://t.co/a7ijKBRihz
https://t.co/XW8BhcRyJ4 https://t.co/a7ijKBRihz
How do internet shutdowns intersect with violations of political rights and state violence across Africa and the Middle East? If you're at RightsCon in-person, join our dialogue *today* at 12.45 in the Puntarenas room. Details below ⬇️
February 16, 2025 at 11:27 AM
How do internet shutdowns intersect with violations of political rights and state violence across Africa and the Middle East? If you're at RightsCon in-person, join our dialogue *today* at 12.45 in the Puntarenas room. Details below ⬇️
Glad to be in Nairobi for the 2023 Digital Rights and Inclusion Forum, hosted by @ParadigmHQ. Looking forward to our 2 @GlobalPartnersD sessions tomorrow: 1⃣ Government responses to disinfo in SSA
2⃣ Evading accountability for human rights abuses through internet shutdowns https://t.co/ybiApgoGvE
2⃣ Evading accountability for human rights abuses through internet shutdowns https://t.co/ybiApgoGvE
February 16, 2025 at 11:27 AM
Glad to be in Nairobi for the 2023 Digital Rights and Inclusion Forum, hosted by @ParadigmHQ. Looking forward to our 2 @GlobalPartnersD sessions tomorrow: 1⃣ Government responses to disinfo in SSA
2⃣ Evading accountability for human rights abuses through internet shutdowns https://t.co/ybiApgoGvE
2⃣ Evading accountability for human rights abuses through internet shutdowns https://t.co/ybiApgoGvE
The UK's Online Safety Bill is fundamentally flawed in seeking to create an entirely separate online world for children, argue @ellenejudson of @Demos and @kyletaylor of @fairvoteuk. They discuss the risks of such an approach in @politicshome https://t.co/yF6KYth1kl
February 16, 2025 at 11:27 AM
The UK's Online Safety Bill is fundamentally flawed in seeking to create an entirely separate online world for children, argue @ellenejudson of @Demos and @kyletaylor of @fairvoteuk. They discuss the risks of such an approach in @politicshome https://t.co/yF6KYth1kl
Peggy Hicks of @UNHumanRights shares some excellent thoughts on "Defining the way forward" on @UNESCO's guidelines for digital platform regulation #InternetForTrust:
1️⃣ complex problems don't have simple solutions - presenting simple quick fix answers is unhelpful
1️⃣ complex problems don't have simple solutions - presenting simple quick fix answers is unhelpful
February 16, 2025 at 11:26 AM
Peggy Hicks of @UNHumanRights shares some excellent thoughts on "Defining the way forward" on @UNESCO's guidelines for digital platform regulation #InternetForTrust:
1️⃣ complex problems don't have simple solutions - presenting simple quick fix answers is unhelpful
1️⃣ complex problems don't have simple solutions - presenting simple quick fix answers is unhelpful
"Multilingualism on digital platforms is absolutely essential for user empowerment". Dilara Begum speaking on the panel on User empowerment and complaints mechanisms at @unescoNOW's #InternetForTrust with HRH @PrincessRymAli @eSafetyOffice
@amalia_toledo @CalGish @DamarJuniarto
@amalia_toledo @CalGish @DamarJuniarto
February 16, 2025 at 11:26 AM
"Multilingualism on digital platforms is absolutely essential for user empowerment". Dilara Begum speaking on the panel on User empowerment and complaints mechanisms at @unescoNOW's #InternetForTrust with HRH @PrincessRymAli @eSafetyOffice
@amalia_toledo @CalGish @DamarJuniarto
@amalia_toledo @CalGish @DamarJuniarto
From "Amplification and Its Discontents" by @daphnehk @knightcolumbia 👏👏 https://x.com/daphnehk/status/1615020403256164352
February 16, 2025 at 11:26 AM
From "Amplification and Its Discontents" by @daphnehk @knightcolumbia 👏👏 https://x.com/daphnehk/status/1615020403256164352
Read the latest newsletter by AlgorithmWatch https://r.algorithmwatch.org/nl3/21f8gzjqud4-SLZEdp95Ng
February 16, 2025 at 11:26 AM
Read the latest newsletter by AlgorithmWatch https://r.algorithmwatch.org/nl3/21f8gzjqud4-SLZEdp95Ng
How can we tackle disinformation in a rights respecting way? @SimoneToussi says media literacy initiatives and fact checking are vital but not sufficient to tackle the problem alone - more multistakeholder dialogue is still needed.
February 16, 2025 at 11:25 AM
How can we tackle disinformation in a rights respecting way? @SimoneToussi says media literacy initiatives and fact checking are vital but not sufficient to tackle the problem alone - more multistakeholder dialogue is still needed.
At 12:55 GMT, I'll speaking in our #IGF2022 Lightning Talk on govt responses to online disinfo in Sub-Saharan Africa. Come and join to find out what @GlobalPartnersD, @CHR_HumanRights, @cipesaug, @article19wafric and @PROTEGEQV have been working on! https://t.co/fnIdCWE9Az
February 16, 2025 at 11:25 AM
At 12:55 GMT, I'll speaking in our #IGF2022 Lightning Talk on govt responses to online disinfo in Sub-Saharan Africa. Come and join to find out what @GlobalPartnersD, @CHR_HumanRights, @cipesaug, @article19wafric and @PROTEGEQV have been working on! https://t.co/fnIdCWE9Az
Fascinating paper - personalized content ranking algorithms can be seen as "crystals" which reflect multifaceted dimensions of the self. These "crystals" are dynamic, refined by both our organic use of the service as well as our intentional efforts to train them. https://t.co/QPO0Ispswl
February 16, 2025 at 11:22 AM
Fascinating paper - personalized content ranking algorithms can be seen as "crystals" which reflect multifaceted dimensions of the self. These "crystals" are dynamic, refined by both our organic use of the service as well as our intentional efforts to train them. https://t.co/QPO0Ispswl
In the introduction, @hwasser highlights the tendency for repressive governments to use laws and policies criminalising disinformation in order to suppress free speech. Our new tool #LEXOTA tracks and analyses these responses across the SSA region: see http://www.lexota.org.
February 16, 2025 at 11:22 AM
In the introduction, @hwasser highlights the tendency for repressive governments to use laws and policies criminalising disinformation in order to suppress free speech. Our new tool #LEXOTA tracks and analyses these responses across the SSA region: see http://www.lexota.org.
Great reflections from Dr Ellie Sakhaee of @Microsoft in @AllTechIsHuman's AI and Human Rights Report
February 16, 2025 at 11:22 AM
Great reflections from Dr Ellie Sakhaee of @Microsoft in @AllTechIsHuman's AI and Human Rights Report
Looking forward to digging into this very comprehensive research by @RIAnetwork, @internetlabbr, @LIRNEasia and @ARIJNetwork on 'Meeting the challenges of information disorder in the Global South'
https://idl-bnc-idrc.dspacedirect.org/handle/10625/60954
https://idl-bnc-idrc.dspacedirect.org/handle/10625/60954
February 16, 2025 at 11:22 AM
Looking forward to digging into this very comprehensive research by @RIAnetwork, @internetlabbr, @LIRNEasia and @ARIJNetwork on 'Meeting the challenges of information disorder in the Global South'
https://idl-bnc-idrc.dspacedirect.org/handle/10625/60954
https://idl-bnc-idrc.dspacedirect.org/handle/10625/60954
Catching up on the @rightscon session on human rights in the metaverse. @brittanheller of the @AtlanticCouncil comments on three particular human rights risks that must be addressed in metaverse apps:
1⃣harassment
2⃣mental privacy 3⃣monetisation and ads
https://t.co/a9QiUPaynU
1⃣harassment
2⃣mental privacy 3⃣monetisation and ads
https://t.co/a9QiUPaynU
February 16, 2025 at 11:21 AM
Catching up on the @rightscon session on human rights in the metaverse. @brittanheller of the @AtlanticCouncil comments on three particular human rights risks that must be addressed in metaverse apps:
1⃣harassment
2⃣mental privacy 3⃣monetisation and ads
https://t.co/a9QiUPaynU
1⃣harassment
2⃣mental privacy 3⃣monetisation and ads
https://t.co/a9QiUPaynU
V. helpful graph in @article19org toolkit on legal thresholds for types of hate speech. A19 note that whilst the hate speech that is "lawful" under A19(2) of ICCPR can't be banned, its link to intolerance still merits a critical response by states https://t.co/xtMnxXj3U4
February 16, 2025 at 11:20 AM
V. helpful graph in @article19org toolkit on legal thresholds for types of hate speech. A19 note that whilst the hate speech that is "lawful" under A19(2) of ICCPR can't be banned, its link to intolerance still merits a critical response by states https://t.co/xtMnxXj3U4