And we need to get clear about this, because LLMs •do• embed bias and their designers •can• steer and shape that bias and people •will• and •are• using that to shape public thought. Closed-source LLMs are an oligarch’s wet dream of thought control. And not all the oligarchs are as clueless as Musk.
We need to get clear on what kinds of evidence show bias — and bias by design — in LLMs. We need to train reporters in how to report this. We need to get our tools sharp. Like…yesterday.
/end
Hearing the Order: Investigating Selection Bias in Large Audio-Language Models
Yu-Xiang Lin, Chen-An Li, Sheng-Lun Wei, Po-Chun Chen, Hsin-Hsi Chen, Hung-yi Lee
https://arxiv.org/abs/2510.00628
Impact of projection-induced optical selection bias on the weak lensing mass calibration of galaxy clusters
Titus Nyarko Nde, Hao-Yi Wu, Shulei Cao, Gladys Muthoni Kamau, Andrius Tamosiunas, Chun-Hao To, Conghao Zhou
https://arxiv.org/abs/2510.00753
Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM Diversity
Jiayi Zhang, Simon Yu, Derek Chong, Anthony Sicilia, Michael R. Tomz, Christopher D. Manning, Weiyan Shi
https://arxiv.org/abs/2510.01171
An LLM generates a statistically typical response.
If you ask an LLM whether the world is round, it generates what would be a typical response to the question of whether the world is round. (It will probably be correct, because the typical answer to that question is the correct one.)
If you ask an LLM to provide examples of its own bias, it will generate what would be a typical response to a request to provide examples of bias.
In neither case is the LLM sharing any knowledge or insight other than what is typical.
2/
How can the use of different modes of survey data collection introduce bias? A simple introduction to mode effects using directed acyclic graphs (DAGs)
Georgia D Tomova, Richard J Silverwood, Peter WG Tennant, Liam Wright
https://arxiv.org/abs/2510.00900
Panorama: misogyny and racism in the metropolitan police.
I'm waiting for the panorama report that highlights excessive bias for promoting Nick Griffin/BNP, Nigel Farage/UKIP, Reform; Boris Johnson/Tory and suppressing the comments of anyone to the left of Labour.
TabINR: An Implicit Neural Representation Framework for Tabular Data Imputation
Vincent Ochs, Florentin Bieder, Sidaty el Hadramy, Paul Friedrich, Stephanie Taha-Mehlitz, Anas Taha, Philippe C. Cattin
https://arxiv.org/abs/2510.01136
The Hidden Bias That Controls Every Decision You Make
https://www.youtube.com/watch?v=jVAl9qHb3FI
What irks me about this is that Grok •is• quite obviously designed and built to exhibit bias; only fools and trolls would claim otherwise.
It’s just…what this article reports isn’t evidence of anything. An LLM can’t “admit” anything. This is nonsense reporting. It’s clickbait. It’s garbage.
3/
The Multivariate SEM-PGS Model: Using Polygenic Scores to Investigate Cross-Trait Genetic Nurture and Assortative Mating
Xuanyu Lyu, Jared Balbona, Tong Chen, Matthew C. Keller
https://arxiv.org/abs/2510.00353
Here is a White House website falsely asserting that military personnel CANNOT refuse illegal orders
And falsely accusing Democrats of asserting Trump has issued illegal orders -- in a video which doesn't mention Trump
The felon doth protest too much
https://www.whitehouse.gov/mediabias/
The Clustering of Active Galactic Nuclei and Star Forming Galaxies in the LoTSS DeepFields
C. L. Hale, P. N. Best, K. J. Duncan, R. Kondapally, M. J. Jarvis, M. Magliocchetti, H. J. A. R\"ottgering, D. J. Schwarz, D. J. B. Smith, J. Zheng
https://arxiv.org/abs/2510.01029
Anthropic open sources a method to score AI model political evenhandedness; Gemini 2.5 Pro got 97%, Grok 4 96%, Claude Opus 4.1 95%, GPT-5 89%, and Llama 4 66% (Ina Fried/Axios)
https://www.axios.com/2025/11/13/anthropic-bot-bias-data
Könnte confirmation bias sein, aber irgendwie häuft sich bei mir der Eindruck, dass Software schlechter wird. Excel ist langsamer als früher, wenn es um große Tabellen oder Autovervollständigung geht. Die App der Badischen Zeitung muss häufiger neu gestartet werden. Pixums Fotobuch-Software vergisst regelmäßig die Einstellung für das Grid. Usw.
#SteadySupporter
Wie neutral sind KI-Systeme bei der medizinischen Diagnostik?
Eine neue Studie zeigt, dass selbst auf Expertenniveau arbeitende Systeme #Krankheiten übersehen und dabei marginalisierte Gruppen benachteiligen können.
Writing unit tests for my random number generation library continues to be difficult. My tests are failing because the bias in the distribution exceeds my expectations, but I'm wondering whether I should just repeat the test more times and permit it to exceed expectations some of the time (as long as it does it symmetrically/rarely/etc. My gut tells me that second-order expectations aren't any better than first-order expectations, but another part of me disagrees.
Thinking more as I write this (writing is thinking): second-order tests can at least give me better info to work with towards fixing things I think! So maybe I'll invest in them.
#coding
Glad to hear the Director of the BBC is resigning in disgrace at their utter failure to ensure a 2026 series of Doctor Who. Any DG that can't even arrange for ten episodes of Doctor Who is clearly incompetent and unsuited to the...
Wait? It was nothing to do with Doctor Who? It was about Political Bias?
They think he was too left wing?
Hahhhahaa. Hahhaa. 🤣 😆
Too left wing.
😆
Stop it, you're killing me.
🤣
#bbc #bias
An investigation finds rampant caste bias in ChatGPT and Sora; a researcher also finds caste bias in Sarvam AI, which touts itself as a sovereign AI for India (Nilesh Christopher/MIT Technology Review)
https://www.technologyreview.com/2025/10/01/1124621/openai-in…
@… @… yeah, I always feel awkward boosting things that might give them an 11ty bias but someone in the community might want to pitch in!
Alarming findings were published in the journal "Cell Reports Medicine",
showing that four leading
AI-enhanced pathology diagnostic systems
differ in accuracy depending on patients‘ age, gender, and race
— demographic data that the AI is extracting directly from pathology slides ... a feat that’s impossible for human doctors.
What is bias but the broken autoimmune response of cisheteropatriarchy?
The Social Cost of Intelligence: Emergence, Propagation, and Amplification of Stereotypical Bias in Multi-Agent Systems
Thi-Nhung Nguyen, Linhao Luo, Thuy-Trang Vu, Dinh Phung
https://arxiv.org/abs/2510.10943
"I do not want to sound like a tin-foil hat conspiracy theorist, but your country, your neighbourhood, your job…your entire life is already controlled by artificial intelligence. If you do not like the bit in Star Trek: Discovery where we find that Control has taken over Starfleet, then I have bad news: an AI is running the military, too."
As beautiful as a sailboat looks. In my eyes it's totally not worth the effort to own one (on lakes in Germany). While I enjoy helping my friends with their boats occasionally ... the amount of maintenance just confirms my bias every time :D
I like to play around as an anonymous commenter in online newspaper columns. If you point out the biases of #AI systems, the comment gets deleted because it is considered too polemical.
The comment was addressing an article about AI in public service and the use in refugee applications 🫣
The toy aisle is still full of gender bias. Here's how to navigate it these holidays
https://phys.org/news/2025-12-toy-aisle-full-gender-bias.html
Remote coding interviews where you have to use your own machine have so much implicit bias built into them. As I haven't had a personal OS X machine in a decade, I have to use my Linux box which leads to...interesting shenanigans.
Ein Artikel in Plos One beleuchtet den politischen #Bias von #Chatbots und zeigt, dass viele KI-Modelle eher linksgerichtete Antworten geben.
Dies könnte auf das Supervised Fine-Tuning zurückzuführen sein, bei dem KI durch menschliche Beispiele lernt. Interessanterweise weisen unterschiedliche
But in all seriousness, take it with a pinch of salt and check with the vendor directly if you really care. Checking this stuff is hard.
Honestly, at best I can really only say, "I *think* the others have not updated but I may be wrong."
Also expect bias when I talk about competitors [and expect the same from representatives of other browsers].
I love Cosmology. Nothing is more worthless than last year's textbook. Especially on the basics. It keeps the field humble.
Strong progenitor age bias in supernova cosmology – II. Alignment with DESI BAO and signs of a non-accelerating universe | Monthly Notices of the Royal Astronomical Society | Oxford Academic
https://academic.oup.com/mnras/article/544/1/975/8281988
A profile of Wikipedia co-founder Larry Sanger, who left in 2002 and whose claims that Wikipedia has a liberal bias are fueling the right's campaign against it (Will Oremus/Washington Post)
RE: https://neuromatch.social/@MolemanPeter/115604657740821359
Survivor bias in this case is: only men that survived after screening report this and not the men that died despite the screening (sorry for the rather cynical way of wording). In additi…
The Oatmeal comic about "AI" "art" is great, but I have to critize the "I like AI, I use AI" part; there's a selection bias at work—he sees the generated "art" as very obviously for what it is, (dehumanizing low-effort slop) but doesn't make the logical conclusion that this is true for _any type of output_ (e.g. text); it always dehumanizing low-effort slop that's not worth looking at, listening to or reading.
"A new investigation of Elon #Musk’s X by Sky News found that every account set up by reporters, 'no matter their political orientation, was fed a glut of rightwing content', much of which was extreme. The experts it consulted believe this pattern could have resulted only from an algorithm engineered for this purpose, and that 'an algorithmic bias must be decided by senior people at …
Senator Marsha Blackburn said
“I look forward to introducing the
"TRUMP AMERICA AI Act"
in the new year to create one federal rulebook for AI
to protect children, creators, conservatives, and communities"
Combats the consistent pattern of bias against conservative figures demonstrated by Big Tech and AI systems by requiring:
Audits of high-risk AI systems to undergo regular bias evaluations to prevent discrimination based on protected characte…
Wikipedia’s rejoinder to Ted Cruz is entirely fair and accurate, but don’t kid yourself that Cruz is interested in fairness or accuracy, he’s just pushing back against the tendency of factual reporting to exhibit a liberal bias.
#wikipedia
AI is Dunning-Kruger as a service
«A cognitive bias, where people with little expertise or ability assume they have superior expertise or ability. This overestimation occurs as a result of the fact that they don’t have enough knowledge to know they don’t have enough knowledge.»
🤯 https://…
GAMBIT : A Challenge Set for Evaluating Gender Bias in Machine Translation Quality Estimation Metrics
Giorgos Filandrianos, Orfeas Menis Mastromichalakis, Wafaa Mohammed, Giuseppe Attanasio, Chrysoula Zerva
https://arxiv.org/abs/2510.06841
Strong progenitor age bias in supernova #cosmology – I. Robust and ubiquitous evidence from a larger sample of host galaxies in a broader redshift range / II. Alignment with DESI BAO and signs of a non-accelerating universe: https://academic.oup.com/mnras/article/538/4/3340/8098234 / https://academic.oup.com/mnras/article/544/1/975/8281988 -> Universe's expansion 'is now slowing, not speeding up': https://ras.ac.uk/news-and-press/research-highlights/universes-expansion-now-slowing-not-speeding (this has been claimed by the group since at least 2023 - http://vietnam.in2p3.fr/2023/windows/transparencies/02_tuesday/Astro_5/01_Son.pdf - and is rejected by the SN cosmology community as https://bsky.app/profile/astromarc.bsky.social/post/3m4xmikwfds26 explains).
Bias-Aware AI Chatbot for Engineering Advising at the University of Maryland A. James Clark School of Engineering
Prarthana P. Kartholy, Thandi M. Labor, Neil N. Panchal, Sean H. Wang, Hillary N. Owusu
https://arxiv.org/abs/2510.09636
House Republicans held a hearing Wednesday broadcasting long-standing conservative allegations of a left-wing bias in the small, prestigious Truman Scholarship program.
Witnesses called by the GOP said the winners disproportionately espouse causes such as promoting racial justice and fighting climate change
—and wind up working for Democrats and left-leaning organizations
—while few recipients profess interest in conservative aims.
But rather than counter the allegations…
Not in Sync: Unveiling Temporal Bias in Audio Chat Models
Jiayu Yao, Shenghua Liu, Yiwei Wang, Rundong Cheng, Lingrui Mei, Baolong Bi, Zhen Xiong, Xueqi Cheng
https://arxiv.org/abs/2510.12185
BiasBusters: Uncovering and Mitigating Tool Selection Bias in Large Language Models
Thierry Blankenstein, Jialin Yu, Zixuan Li, Vassilis Plachouras, Sunando Sengupta, Philip Torr, Yarin Gal, Alasdair Paren, Adel Bibi
https://arxiv.org/abs/2510.00307
Days after immigration officials announced the death of a detainee at Newark migrant jail Delaney Hall,
Rep. LaMonica McIver visited the facility for the first time since prosecutors allege she assaulted federal agents there during a melee in May.
McIver’s visit coincided with a new filing in the Newark Democrat’s criminal case
in which her attorneys accuse prosecutors of failing to produce key evidence that they say shows bias among the agents involved in the scuffle,
Does LLM Focus on the Right Words? Diagnosing Language Bias in LLM-based Recommenders
Bohao Wang, Jiawei Chen, Feng Liu, Changwang Zhang, Jun Wang, Canghong Jin, Chun Chen, Can Wang
https://arxiv.org/abs/2510.10978
Mitigating Judgment Preference Bias in Large Language Models through Group-Based Polling
Shuliang Liu, Zhipeng Xu, Zhenghao Liu, Yukun Yan, Minghe Yu, Yu Gu, Chong Chen, Huiyuan Xie, Ge Yu
https://arxiv.org/abs/2510.08145
Microsoft signs a deal with a legal advocacy group to resolve claims it denied discounts to conservative religious nonprofits; OpenAI and Asana also made deals (Jeff Green/Bloomberg)
https://www.bloomberg.com/news/articles/20
Trump’s White House is taking on the role of media critic and asking for help from “everyday Americans.”
Over the Thanksgiving weekend, the White House launched a web portal it says will spotlight bias on the part of news outlets,
targeting the Boston Globe, CBS News and the Independent as its inaugural “media offender of the week.”
It’s the latest wrinkle in the fight against the First Amendment
-- what Trump, in his first term, labeled “fake news.”
The Republi…
From Noisy Traces to Stable Gradients: Bias-Variance Optimized Preference Optimization for Aligning Large Reasoning Models
Mingkang Zhu, Xi Chen, Bei Yu, Hengshuang Zhao, Jiaya Jia
https://arxiv.org/abs/2510.05095
OpenAI says GPT‑5 instant and GPT‑5 thinking cut political bias by 30% from earlier models, and show greater robustness to charged prompts (Ashley Gold/Axios)
https://www.axios.com/2025/10/09/openai-gpt-5-least-biased-model
IndiCASA: A Dataset and Bias Evaluation Framework in LLMs Using Contrastive Embedding Similarity in the Indian Context
Santhosh G S, Akshay Govind S, Gokul S Krishnan, Balaraman Ravindran, Sriraam Natarajan
https://arxiv.org/abs/2510.02742
Spatially-informed transformers: Injecting geostatistical covariance biases into self-attention for spatio-temporal forecasting
Yuri Calleo
https://arxiv.org/abs/2512.17696 https://arxiv.org/pdf/2512.17696 https://arxiv.org/html/2512.17696
arXiv:2512.17696v1 Announce Type: new
Abstract: The modeling of high-dimensional spatio-temporal processes presents a fundamental dichotomy between the probabilistic rigor of classical geostatistics and the flexible, high-capacity representations of deep learning. While Gaussian processes offer theoretical consistency and exact uncertainty quantification, their prohibitive computational scaling renders them impractical for massive sensor networks. Conversely, modern transformer architectures excel at sequence modeling but inherently lack a geometric inductive bias, treating spatial sensors as permutation-invariant tokens without a native understanding of distance. In this work, we propose a spatially-informed transformer, a hybrid architecture that injects a geostatistical inductive bias directly into the self-attention mechanism via a learnable covariance kernel. By formally decomposing the attention structure into a stationary physical prior and a non-stationary data-driven residual, we impose a soft topological constraint that favors spatially proximal interactions while retaining the capacity to model complex dynamics. We demonstrate the phenomenon of ``Deep Variography'', where the network successfully recovers the true spatial decay parameters of the underlying process end-to-end via backpropagation. Extensive experiments on synthetic Gaussian random fields and real-world traffic benchmarks confirm that our method outperforms state-of-the-art graph neural networks. Furthermore, rigorous statistical validation confirms that the proposed method delivers not only superior predictive accuracy but also well-calibrated probabilistic forecasts, effectively bridging the gap between physics-aware modeling and data-driven learning.
toXiv_bot_toot
Sony unveils the Fair Human-Centric Image Benchmark dataset to test the fairness of computer vision models, saying it was compiled in a fair and ethical manner (Thomas Claburn/The Register)
https://www.theregister.com/2025/11/05/sony_ai_vision_model_benchmark/…
How ideology-driven AI chatbots like Grok and Gab's Arya position themselves as alternatives to mainstream chatbots accused of liberal bias (New York Times)
https://www.nytimes.com/2025/11/04/busines
@inthehands@hachyderm.io
