OpenAI’s #Sora Is Plagued by #Sexist , #Racist , and #Ableist #Biases
WIRED tested the popular #AI video generator from #OpenAI and found that it amplifies sexist #stereotypes and ableist tropes, perpetuating the same biases already present in AI image tools.
https://www.wired.com/story/openai-sora-video-generator-bias/
Does anybody has a simple (set of) schema for explaining how LLM AI is built, from data labelling and model training to saas and system prompt and censorship guards.
Context: my son is doing a school project where he wants to identify different sources of bias in generative AI, and so if he had a schema with the different steps, it would be easier.
I wasn't able to find something high level enough for that.
In French for bonus points :))
Repost is 
(now online) Training #ML on fair data is key to ensure the resulting ML models will be from #biases and other #ethical concerns. 
But is scientific data annotated with enough metadata to evaluate potential #biases in the data (and its gathering process)? (TL;DR it's a disaster 
)
Are at least improving on this? What can we do to fix this situation? 
https://www.nature.com/articles/s41597-025-04402-4
(work led by Joan Giner Miguelez with Abel Gómez and yours truly)
#openaccess hashtag#scientificdata #empirical hashtag#fairness #research #machinelearning #trustworthiness #ResponsibleAI
You have to greet a group of strangers in a casual setting. Which group + greeting combo(s) would you be most uncomfortable using?
For group purposes, consider a femme group to include any combination cis/trans people who identify as women; masc group as men; and mixed group as men, women, and/or nonbinary folx.
For greeting purposes, consider "guys" to be interchangeable with any other masculine collective term (i.e. boys, men, etc.), and "gals" to be interchangeable with any other feminine collective term (i.e. girls, ladies, etc.).
#accessible
#activism
#activists
#advocacy
#advocate
#advocates
#affirming_care
#all_inclusive
#allyship
#anti_racism
#antiracist
#assigned_at_birth
#assigned_female_at_birth
#assigned_male_at_birth
#at_risk
#barrier
#barriers
#belong
#bias
#biased
#biased_toward
#biases
#biases_towards
#biologically_female
#biologically_male
#BIPOC
#Black
#breastfeed_people
#breastfeed_person
#chestfeed_people
#chestfeed_person
#clean_energy
#climate_crisis
#climate_science
#commercial_sex_worker
#community_diversity
#community_equity
#confirmation_bias
#cultural_competence
#cultural_differences
#cultural_heritage
#cultural_sensitivity
#culturally_appropriate
#culturally_responsive
#DEI
#DEIA
#DEIAB
#DEIJ
#disabilities
#disability
What Happened to #JonStewart? — A Retrospective
This is an already 2-years old video, but I would recommend it strongly to everyone interested in understanding the role of #politicalsatire in nowadays #public #discourses, and especially to fans of the #DailyShow with Jon Stewart, as they are particularly likely to share some of his most problematic #biases, that is the appeal to #rationality and this idea that only #polite conversation is fit for #democracy.
@jerome Yes. It is a product of the #biases of the culture in which it was developed. The ONLY way for clean and transparent AI/LLM models to develop and not become tools of the #oligarchy #government, is from #Competition. I keep using many different AI models from many different cultures and countries. It's the only way.
I'm sure you use #LLMs for a text-to-image generation? 
BUT, are you aware of all the #social #biases that those images may embed?
In our latest paper "Evaluating Representational Harms in Text-to-Image Models" we present a #DSL and #testing #framework to #test #ethical concerns in image #generation from #prompts.
https://modeling-languages.com/evaluating-representational-harms-in-text-to-image-models/ #openaccess
https://github.com/SOM-Research/ImageBiTe
Work to be presented at CAIN: International Conference on AI Engineering - Software Engineering for AI .
#Scientific papers include more and more often #replication packages with valuable #datasets that are then frequently used to train #AI and #MachineLearing models.
But with datasets not properly #documented with information about their #provenance, #biases and other #social concerns, this is risky as the #ML models will use in environments for which the data was not representative, yielding potentially wrong conclusions.
In this work, we have analyzed the datasets in two top dataset journals to study their #documentation #practices and propose a few recommendations to improve the current situation.
Paper accepted in the 𝘕𝘢𝘵𝘶𝘳𝘦'𝘴 𝘚𝘤𝘪𝘦𝘯𝘵𝘪𝘧𝘪𝘤 𝘋𝘢𝘵𝘢 journal
Pre-print https://arxiv.org/pdf/2401.10304
Tonight I revisited the chapter on biases in my book. I had the focus entirely wrong. I had an idea so I went back in to see what I could do. Now I absolutely love it!
What a great feeling this is. I’m confident it will really improve the flow and set a stronger foundation for the rest of the book. 
(For those who don’t know, my book is about exclusivity and biases in design.)
Quiet extinctions of invertebrates are largely unnoticed and unmourned
‘This is the way the world ends; not with a bang but a whimper’ TS Eliot, The Hollow Men
"Australian species have become extinct over this 236-year period...We predict that 39–148 species will become extinct in 2024. This is inconsistent with a recent pledge by the Australian government to prevent all extinction...Unless there is a major increase in investment and change in conservation priorities, and more effective control of threats, this rate of extinction will increase. We should not simply maintain the current conservation status quo and let these extinctions happen."
"So long as invertebrate extinctions remain nameless and invisible, this failure cannot be demonstrated, or readily overcome; and efforts will instead be directed towards the less imperilled, but better-known and iconic species."
>>
Woinarski JCZ, Braby MF, Gibb H, et al. This is the way the world ends; not with a bang but a whimper: Estimating the number and ongoing rate of extinctions of Australian non-marine invertebrates. Cambridge Prisms: Extinction. 2024;2:e23. doi:10.1017/ext.2024.26
https://www.cambridge.org/core/journals/cambridge-prisms-extinction/article/this-is-the-way-the-world-ends-not-with-a-bang-but-a-whimper-estimating-the-number-and-ongoing-rate-of-extinctions-of-australian-nonmarine-invertebrates/D0DCAA03EBA7ACC25F98F7BF5D87A2A6
#invertebrates #extinction #narrative #QuietExtinctions #ecosystem #biodiversity #HabitatDestruction #habitat #degradation #taxonomy #TaxonomicInequality #biases #cute #iconic #disinterest #BushBlitz #conservation #values
I've noticed something, news #media is manipulative. Many headlines and articles from mainstream news outlets are written in a deliberately provocative way that reinforces unfounded #biases or #assumptions. #polarization weakens the social bonds that hold communities together. focusing on what divides people you weaken trust. This feels like a deliberate and systematic attack intended to break communities apart and sow distrust. and it seems to be happening independent of political orientation.
“
#ArtificialIntelligence systems tend to take on human #Biases and amplify them, causing people who use that #AI to become more biased themselves, finds a new study …”
https://www.sciencedaily.com/releases/2024/12/241218132137.htm
#PoliceChief begs #transphobes to stop falsely claiming #Wisconsin #schoolshooter was #trans.
"I wish people would leave their own #personal #biases out of this," the local police chief said.
#Women #Transgender #LGBTQ #LGBTQIA #Washington #Conservatives #Extremism #Fascism #Disinformation #Hate #Bigotry #Violence #Transphobia
2. #CriticalThinking: The capacity to analyze and evaluate ideas and policies #EvidenceBased, logic, and reason rather than #Ideological #Biases. It involves questioning assumptions and being willing to change views based on new information.
I wrote a piece about how we learn racism. It’s complicated. But if we acknowledge it instead of running, we can unlearn and do better.
“Most white people struggle with the idea that systemic racism even exists, or that we are personally responsible for fueling it. It stings when you consciously see yourself as an ally, and then find yourself labeled an oppressor. It’s highly uncomfortable. So people avoid it.”
https://markwrites.io/rodney-the-drinking-fountain-and-the-little-white-racist-how-i-learned-to
Flying to London to attend the #Open #Innovation hashtag#AI #Research #Community Annual Research #Workshop where I'll be presenting our #redteaming efforts to uncover #biases in #LLMs.
To know more about our #ethical #testing #framework:https://lnkd.in/eedS2DbD https://lnkd.in/eW5S7aax
All this is part of our efforts in the BESSER project to make sure we can use #lowcode tools to automatically generate #software that it is #ethicalbydesign
I’m currently writing a book. A publisher reached out to me with some unsettling words:
“The problem doesn’t strike me as a real at all, and if it is real, it doesn’t strike me as a high-priority problem. I can think of 25 other problems that are more urgent and interesting to me.”
I’m writing about is exclusivity and biases in design. Every marginalized designer knows it’s very real.
I’m now more encouraged than ever to finish this. 
#GameTheory Research Suggests We Can’t Ever Fully Learn From Our Mistakes : Medium
Can We Keep Our #Biases from Creeping into #AI? : HBR
Norway's #Dragon's Eye: The fantastical 'pothole' that emerged from ice 16,000 years ago : Live Sci
Check our latest #KnowledgeLinks
🅻🅸🅲🅴 (
) 
Klabauter (oald) 
#woke 
