We present a reproduction study of "Competition of Mechanisms: Tracing How Language Models Handle Facts and Counterfactuals" (Ortu et al., 2024), which investigates competition of mechanisms in language models between factual recall and counterfactual in-context repetition. Our study successfully reproduces their primary findings regarding the localization of factual and counterfactual information, the dominance of attention blocks in mechanism competition, and the specialization of attention heads in handling competing information. We reproduce their results on both GPT-2 (Radford et al., 2019) and Pythia 6.9B (Biderman et al., 2023). We extend their work in three significant directions. First, we explore the generalizability of these findings to even larger models by replicating the experiments on Llama 3.1 8B (Grattafiori et al., 2024), discovering greatly reduced attention head specialization. Second, we investigate the impact of prompt structure by introducing variations where we avoid repeating the counterfactual statement verbatim or we change the premise word, observing a marked decrease in the logit for the counterfactual token. Finally, we test the validity of the authors' claims for prompts of specific domains, discovering that certain categories of prompts skew the results by providing the factual prediction token as part of the subject of the sentence. Overall, we find that the attention head ablation proposed in Ortu et al. (2024) is ineffective for domains that are underrepresented in their dataset, and that the effectiveness varies based on model architecture, prompt structure, domain and task.

Over the past few weeks I have been using GPT-4 to help me cook. Need a substitute for an ingredient you forgot to buy? GPT can suggest an alternative. Time to clear out the cupboards? Simply type: "Please create a recipe using two eggs, a jar of borlotti beans, a potato, a leek, and the scrapings on the bottom of a jar of pickle." I'm always polite, and so is GPT. It thinks for a moment – then whips up the instructions for an unusual but edible hash and even wishes me bon appétit.

Most questionnaires offer ordered responses whose order is poorly studied via belief functions. In this paper, we study the consequences of a frame of discernment consisting of ordered elements on belief functions. This leads us to redefine the power space and the union of ordered elements for the disjunctive combination. We also study distances on ordered elements and their use. In particular, from a membership function, we redefine the cardinality of the intersection of ordered elements, considering them fuzzy. Keywords: ordinal variable ordered frame of discernment ordered and fuzzy elements ordered power set distance.

For the 3rd episode of our interview series with AI experts, we had a great conversation with Anna Bethke! Anna Bethke is a Principal Data Scientist focused on fair, accountable, transparent, &…

Artificial Intelligence (AI) is disrupting businesses and job roles in every industry, causing concerns about long-term job security for low-skill manual jobs and management roles alike. To prepare for this AI-driven economy, many experienced managers and seasoned executives are turning to MOOCs (Massive Open Online Courses) to upskill in foundational data analytics and AI. This trend is unlikely to slow down anytime soon: The global MOOC market is expected to grow from $3.9 billion in 2018 to $20.8 billion by 2023, a CAGR of 40.1 percent. Business and technology-related courses make up 40 percent of these online courses. Many universities have also joined the drive to fill the AI leadership gap by offering high-touch executive education programs.

AI IoT: Two significant trends that are dominating the technology industry are the Internet of Things (IoT) and Artificial Intelligence (AI). However, for industrial automation, these two technologies are much more than buzzwords or trending topics. The convergence of AI IoT is going to redefine the future of industrial automation for the Industry 4.0 revolution. IoT and AI are two discrete technologies that have a significant impact on multiple industry verticals. While IoT is the digital nervous system, AI becomes the brain that makes decisions which control the overall system.

Removing a particular piece of data that contributed to a machine learning model is like trying to remove the second spoonful of sugar from a cup of coffee. The data, by this time, has already become intrinsically linked to many other neurons inside the model. If a data point represents'defining' data that was involved in the earliest, high-dimensional part of the training, then removing it can radically redefine how the model functions, or even require that it be re-trained at some expenditure of time and money. Nonetheless, in Europe at least, Article 17 of the General Data Protection Regulation Act (GDPR) requires that companies remove such user data on request. Since the act was formulated on the understanding that this erasure would be no more than a database'drop' query, the legislation destined to emerge from the Draft EU Artificial Intelligence Act will effectively copy and paste the spirit of GDPR into laws that apply to trained AI systems rather than tabular data.

PHILADELPHIA – As the search for successful Alzheimer's disease drugs remains elusive, experts believe that identifying biomarkers -- early biological signs of the disease -- could be key to solving the treatment conundrum. However, the rapid collection of data from tens of thousands of Alzheimer's patients far exceeds the scientific community's ability to make sense of it. Now, with a $17.8 million grant from the National Institute on Aging at the National Institutes of Health, researchers in the Perelman School of Medicine at the University of Pennsylvania will collaborate with 11 research centers to determine more precise diagnostic biomarkers and drug targets for the disease, which affects nearly 50 million people worldwide. For the project, the teams will apply advanced artificial intelligence (AI) methods to integrate and find patterns in genetic, imaging, and clinical data from over 60,000 Alzheimer's patients -- representing one of the largest and most ambitious research undertakings of its kind. Penn Medicine's Christos Davatzikos, PhD, a professor of Radiology and director of the Center for Biomedical Image Computing and Analytics, and Li Shen, PhD, a professor of Informatics, will serve as two of five co-principal investigators on the five-year project.

The technology behind Autonomous vehicles can surprise you. These vehicles are characterized by not having to deal with human limitations, such as tiredness and inattention. To the delight of many, these machines can park alone, and they do not drive drunk or speak on the phone while driving, like many humans that we know. It is known that human failures cause 94% of traffic accidents, and this innovation is mainly developed to save lives, reducing consistently the fatalities. According to a study from 2015 by the National Highway Traffic Safety Administration (NHTSA), traffic accidents are the most significant cause of death of young people between 15 and 29 years globally, overcoming the victims of AIDS, flu, and dengue together, according to the World Health Organization (WHO).

Once again, businesses are under attack. We are in the midst of "Disruption 2.0," and this time, the attack comes from within. Previously, in the age of Disruption 1.0, startups were spearheading the battle against established businesses. The old corporate world appeared to be defenseless in the face of these fast, agile, tech-savvy, micro, and, at times, well-funded companies. The fuel that powered Disruption 1.0 was the fact that the technology, the talent, and the funding driving startups all existed outside the corporate walls.