Dr Abeba Birhane
LightNews LightNews
banner
abeba.bsky.social
Dr Abeba Birhane
@abeba.bsky.social
49K followers 820 following 3.8K posts
Founder & PI @aial.ie. Assistant Professor of AI, School of Computer Science & Statistics, @tcddublin.bsky.social

AI accountability, AI audits & evaluation, critical data studies. Cognitive scientist by training. Ethiopian in Ireland. She/her
Posts Replies Media Videos
abeba.bsky.social
what a great honour to receive the 25th Irish Tatler Women of the Year Award in the category of Innovation. the optimist in me thinks this marks a turn for recognition of the importance of critical work and accountability in AI

www.businesspost.ie/life-luxury/...
November 18, 2025 at 6:13 PM
abeba.bsky.social
science but could be misleading or incorrect 🤷🏾‍♀️🤷🏾‍♀️

such a shame that academic publication, supposedly the guardian of knowledge production, is giving up what's at it's core - trustworthy, accurate, independent, verifiable knowledge - just so academics & publishers alike can get on the genai bandwagon
November 12, 2025 at 6:26 PM
abeba.bsky.social
flat out lies. these tech ceos take regulators and the masses for fools
“If you want to know how a model will behave, the only way of doing it is to run it,” Mensch also suggested. “You do need to have some empirical testing, what’s happening. Knowing the input data that has been used for training is not going to tell you whether your model is going to behave well in [a healthcare use-case], for instance. You don’t really care about what’s in the training data. You do care about the empirical behaviour of the model. So you don’t need knowledge of the training data. And if you had knowledge of the training data, it wouldn’t even teach you whether the model is going to behave well or not. So this is why I’m saying it’s neither necessary nor sufficient.”
November 12, 2025 at 10:38 AM
abeba.bsky.social
flat out lies. these tech ceos take regulators and the masses for fools
“If you want to know how a model will behave, the only way of doing it is to run it,” Mensch also suggested. “You do need to have some empirical testing, what’s happening. Knowing the input data that has been used for training is not going to tell you whether your model is going to behave well in [a healthcare use-case], for instance. You don’t really care about what’s in the training data. You do care about the empirical behaviour of the model. So you don’t need knowledge of the training data. And if you had knowledge of the training data, it wouldn’t even teach you whether the model is going to behave well or not. So this is why I’m saying it’s neither necessary nor sufficient.”
November 12, 2025 at 10:36 AM
abeba.bsky.social
"the law does not self-enforce. the law does not self-execute.

there are more parking ticket enforcement in Brussels than there are enforcement of the DMA" Cori
November 7, 2025 at 9:02 AM
abeba.bsky.social
amen!! #Mozfest
November 7, 2025 at 8:53 AM
abeba.bsky.social
Three Moves to Stop Broligarchy (Break, Build, Enforce) by
Cori Crider and Robin Berjon
mzf.st/15

hands down one of the most refreshing talks at MozFest
November 7, 2025 at 8:45 AM
abeba.bsky.social
November 6, 2025 at 9:53 PM
abeba.bsky.social
gee, won’t regulators think of the poor multi-billioners profiting out of theft, environmental destruction, and shitty and harmful tech they keep pushing aggressively into the public
In correspondence to Mr O’Callaghan, Meta said the EU’s data laws should be revised to take better account of “economic interests” and private sector innovation. Meta asked for the Government’s support in taking concerns about the “ineffectiveness” of the existing data protection framework to Ireland’s EU commissioner Michael McGrath, who oversees GDPR rules in his role as justice commissioner. Dublin could also flag the company’s concerns to other European governments, Ms Egan suggested. “We would be grateful for your support for these changes, and for your advocacy at European level to achieve them,” the Meta executive wrote. The correspondence between Meta and Mr O’Callaghan was released to The Irish Times in response to a Freedom of Information Act request. A spokeswoman for Meta said the company did not have any comments to add beyond what it had set out in the letter. A Department of Justice spokesman said Ireland supported the commission’s plan to simplify EU data protection laws. Though any changes had to “get the balance right”, by reducing some administrative burden, “while continuing to protect the fundamental rights of citizens”, he said.
October 31, 2025 at 6:19 PM
abeba.bsky.social
there goes the empty big tech rhetoric of `innovation or regulation’
October 31, 2025 at 6:12 PM
abeba.bsky.social
regulations that are put in place to protect individuals from the wealthiest and most powerful corporations are apparently `out of control’
Meta, the company that owns Facebook, Instagram and WhatsApp, has privately tried to convince the Irish Government to lead a pushback against data protection laws at European Union level, correspondence shows. The interpretation by regulators of strict EU rules that protect individuals’ data and privacy is “completely out of control”, Meta’s chief privacy officer and vice-president Erin Egan told Minister for Justice Jim O’Callaghan in a May 23rd letter. The correspondence shows Meta has expanded the scope of its campaign against EU regulations to include the bloc’s General Data Protection Regulation (GDPR), which puts greater obligations on big tech multinationals to safeguard the data of people using their platforms or services. Europe’s approach to data protection requires a total overhaul, Ms Egan told Mr O’Callaghan in her letter.
October 31, 2025 at 6:09 PM
abeba.bsky.social
the tech industry is a threat to scientific knowledge, academic freedom and evidence-driven industry regulation. the industry systematically captures and/or interferes with and undermines independent knowledge production processes that challenge it

arxiv.org/pdf/2510.19894
ABSTRACT Emerging information technologies like social media, search engines, and AI can have a broad impact on public health, political institutions, social dynamics, and the natural world. It is critical to develop a scientific understanding of these impacts to inform evidence-based technology policy that minimizes harm and maximizes benefits. Unlike most other global-scale scientific challenges, however, the data necessary for scientific progress are generated and controlled by the same industry that might be subject to evidence-based regulation. Moreover, technology companies historically have been, and continue to be, a major source of funding for this field. These asymmetries in information and funding raise significant concerns about the potential for undue industry influence on the scientific record. In this Perspective, we explore how technology companies can influence our scientific understanding of their products. We argue that science faces unique challenges in the context of technology research that will require strengthening existing safeguards and constructing wholly new ones.
October 24, 2025 at 4:03 PM
abeba.bsky.social
"under-appreciated fact: large corporations have internal policy teams with 100s of employees whose full-time jobs are to study the global policy environment, advise their companies on political matters, and seek to influence the policy process." www.andrewbenjaminhall.com/HallSun25.pdf
In this paper, we define and measure a previously unstudied channel by which companies react to, and attempt to shape, politics: internal policy teams. We use the text of more than 100,000 job listings to classify the roles of roughly 100 million workers, identifying more than 250,000 individuals working in policy roles in the US. We estimate that policy teams are—very roughly—13 times larger in size than lobbying teams, on average, and are increasing in size relative to lobbyists and other corporate positions over the last two decades. Policy teams and lobbyists appear to be complements; both across firms and within, increases in lobbying predict increases in policy team size. But policy team members are quite different from lobbyists: they are much less likely to have previous government experience, and they are less balanced in terms of their partisanship. Taken together, our findings show that corporations invest far more in politics than previously recognized, and suggest that a significant part of this investment goes towards developing internal expertise and communicating with a broader set of policymakers and policy-adjacent actors, rather than towards quid-pro-quo lobbying
October 24, 2025 at 3:09 PM
abeba.bsky.social
the flaw in this logic is that all commercial llms (by opeAI, google, anthropic etc) hardly share information about their models inc training data, evaluation processes or benchmarks. they basically operate under a “trust me bro” attitude. hence why this analogy is misguided and/or flat out wrong
Of Models and Tin Men- A behavioural economics study of principal-agent problems in AI alignment using large-language models One of the challenges in assessing the capability and safety of LLMs is their opacity; AI safety researchers sometimes refer to them as “giant inscrutable arrays of fractional numbers” [6]. Since many of their capabilities emerge down- stream, we cannot reason about their high-level functioning theoretically from first principles, but rather we have to resort to empirical enquiry and actual experiments (c.f. [7, 8, 9]). This is analogous to using the scientific method to understand human cognition. The human brain is similarly opaque, and therefore behavioural psychologists test their hypotheses using controlled experiments with human subjects. Accordingly, researchers have started to adapt empirical methods from cognitive psychology [10], and behavioural economics [11, 12, 13, 14, 15, 16, 17, 14], to the study of large-language models
October 18, 2025 at 2:47 PM
abeba.bsky.social
brought to you by the same company that is building you “agentic” ai and “agi”
October 15, 2025 at 6:21 PM
abeba.bsky.social
a lot of technical jargon to say surveillance and AI pseudoscience

(from a cfp for an AI conference that somehow landed in my inbox)
Gaze modeling and interpretation, Social network loneliness, Gesture interaction for AR applications Sharing gestures in remote collaboration, AV mirrors Eye Gaze and hand gesture sharing Physiological sensing in VR Individualized VR for mental health Generative-AI for human experience development AI for human digital twins Integration of XR in primary education
October 15, 2025 at 1:55 PM
abeba.bsky.social
i don't have the words to emphasise how seriously bad things are gonna get
October 14, 2025 at 8:18 PM
abeba.bsky.social
🤦🏾‍♀️
in order to align llms with human values, we use llms to base our understanding of human values
reads: Given the financial and logistical challenges linked to human annotation, we choose to employ the feedback generated by the GPT-3.5-turbo API as a substitute for human feedback.
October 13, 2025 at 4:47 PM
abeba.bsky.social
friday vibes
October 10, 2025 at 3:34 PM
abeba.bsky.social
we wrote this over 5 yrs ago

dl.acm.org/doi/abs/10.1...
October 9, 2025 at 8:56 PM
abeba.bsky.social
an ai video of Sam Altman shoplifting but much more interesting what's in the training data. there's no way these models can reimagine camera viewpoints unless a nearest neighbour exists in the training dataset. looks like target video feed must have entered the dataset
October 6, 2025 at 12:39 PM
abeba.bsky.social
sigh
I have attached my CV and academic transcripts for your review. I would be happy to provide any additional information you might need. Thank you for your time and consideration. I look forward to hearing from you. Best regards, [Your Name] [Your Contact Information]
October 6, 2025 at 9:03 AM
abeba.bsky.social
Robot personhood/rights is conceptually bogus and legally puts more power/rights in the hands of those that develop and deploy robots/AI systems

firstmonday.org/ojs/index.ph...
October 5, 2025 at 8:28 PM
abeba.bsky.social
Yesterday the Royal Society held a series of panels & keynotes in celebration of the 75th anniversary of the Turing Test. The unanimous critical takes on AGI, big tech's power concentration & the call for urgent action left me w hope things might be turning

Watch here m.youtube.com/live/GmnBTCK...
October 3, 2025 at 6:38 PM
abeba.bsky.social
saw this at heathrow airport. i can't see it appealing to anyone but fraudsters and scammers
October 3, 2025 at 5:41 PM