Understanding what knowledge is implicitly encoded in deep learning models is essential for improving the interpretability of AI systems. This paper examines common methods to explain the knowledge encoded in word embeddings, which are core elements of large language models (LLMs). These methods typically involve mapping embeddings onto collections of human-interpretable semantic features, known as feature norms. Prior work assumes that accurately predicting these semantic features from the word embeddings implies that the embeddings contain the corresponding knowledge. We challenge this assumption by demonstrating that prediction accuracy alone does not reliably indicate genuine feature-based interpretability. We show that these methods can successfully predict even random information, concluding that the results are predominantly determined by an algorithmic upper bound rather than meaningful semantic representation in the word embeddings. Consequently, comparisons between datasets based solely on prediction performance do not reliably indicate which dataset is better captured by the word embeddings. Our analysis illustrates that such mappings primarily reflect geometric similarity within vector spaces rather than indicating the genuine emergence of semantic properties.

Artificial intelligence can already write poems and make movie recommendations. What's it going to do next? High school students in Longmont, Colorado, are learning how to design their own AI model projects at the St. Vrain Valley School District Innovation Center. The program started this past fall. Mai Vu, the A.I. Program manager at St. Vrain Valley School District, said the AI program's goal is to teach students how to use AI to solve real-world problems.

The Biden Administration will start implementing a new requirement for the developers of major artificial intelligence systems to disclose their safety test results to the government. The White House AI Council is scheduled to meet Monday to review progress made on the executive order that President Joe Biden signed three months ago to manage the fast-evolving technology. Read More: Why Biden's AI Executive Order Only Goes So Far Chief among the 90-day goals from the order was a mandate under the Defense Production Act that AI companies share vital information with the Commerce Department, including safety tests. Ben Buchanan, the White House special adviser on AI, said in an interview that the government wants "to know AI systems are safe before they're released to the public -- the president has been very clear that companies need to meet that bar." The software companies are committed to a set of categories for the safety tests, but companies do not yet have to comply with a common standard on the tests.

The White House says "developers of the most powerful AI systems" will now have to report AI safety test results to the Department of Commerce in the wake of an executive order issued by President Biden aimed at "managing the risks" of the technology. The news comes as Deputy Chief of Staff Bruce Reed is convening the White House AI Council on Monday, consisting of "top officials from a wide range of federal departments and agencies" who have reported completing 90-day actions and advancing other directives tasked by the order Biden signed last October, according to the White House. Among those actions was that they "[u]sed Defense Production Act authorities to compel developers of the most powerful AI systems to report vital information, especially AI safety test results, to the Department of Commerce," the White House said. "These companies now must share this information on the most powerful AI systems, and they must likewise report large computing clusters able to train these systems," the White House added. The White House announced Monday that companies that are working on the "most powerful AI systems" must now report "AI safety test results" to the Department of Commerce.

The growth of computationally intensive technologies such as machine learning incurs a high carbon footprint and is contributing to climate change. Alongside that rapid growth is an expanding portfolio of green AI tools and techniques to help offset carbon usage and provide a more sustainable path forward. The cost to the environment is high, according to research published last month by Microsoft and the Allen Institute for AI, with co-authors from Hebrew University, Carnegie Mellon University and Hugging Face, an AI community. The study extrapolated data to show that one training instance for a single 6 billion parameter transformer ML model -- a large language model -- is the CO2 equivalent to burning all the coal in a large railroad car, according to Will Buchanan, product manager for Azure machine learning at Microsoft, Green Software Foundation member and co-author of the study. In the past, code was optimized in embedded systems that are constrained by limited resources such as those seen in phones, refrigerators or satellites, said Abhijit Sunil, analyst at Forrester Research.

When OpenAI demonstrated a powerful artificial intelligence algorithm capable of generating coherent text last June, its creators warned that the tool could potentially be wielded as a weapon of online misinformation. Now a team of disinformation experts has demonstrated how effectively that algorithm, called GPT-3, could be used to mislead and misinform. The results suggest that although AI may not be a match for the best Russian meme-making operative, it could amplify some forms of deception that would be especially difficult to spot. Over six months, a group at Georgetown University's Center for Security and Emerging Technology used GPT-3 to generate misinformation, including stories around a false narrative, news articles altered to push a bogus perspective, and tweets riffing on particular points of disinformation. "I don't think it's a coincidence that climate change is the new global warming," read a sample tweet composed by GPT-3 that aimed to stoke skepticism about climate change.

News that Pearson, the world's largest textbook publisher, is phasing out print publications for higher education to adopt a resolutely digital-first policy may signal an eventual full stop for traditional book learning. In the view of Mike Buchanan, executive director of HMC, which represents independent school head teachers, digital education will unlock a less rigid approach to classroom-based learning, as well as enable closer collaboration with pupils' families. "In a growing number of schools, the use of modern management information and recording systems to harvest details of classroom activities and pupil progress is already allowing parents to access and aggregate their child's attainment records," he says. "In the future, this will no doubt see the traditional termly report being replaced by daily digital updates." Mr Buchanan predicts individual academic achievement will be charted by artificial intelligence (AI), rather than by a plethora of exams, and argues that for teachers disenchanted by the current need to "teach to the test", the freedom to pursue a more rounded curriculum will foster a new optimism.

This article frames and presents the discussion at an invited panel “AI History: Expert Systems” held at the AAAI-17 conference held in San Francisco. The panel’s purpose was to open up this history of expert systems, its transformational aspects, and its connections to today’s AI awakening.

In all the recent Roseanne micro-controversies, one incredible statement by Roseanne Barr seems to have eluded her critics. "I have 20 fortnite victory royale wins," she tweeted back in February. It's not clear how serious the 65-year-old comeback kid was being. Mostly because, as Fortnite Battle Royale players will know, to have 20 Victory Royale wins is incredibly hard. Gamers start off marooned on an island with a hundred other online players.

In the history of driverless cars, the Defense Advanced Research Projects Agency's 2004 and 2005 Grand Challenges contributed to improving the tech to the point of viability. The government could play a similar role in the development of artificial intelligence, said Jack Clark, the director of Open AI, at an April 18 congressional hearing focusing on government's role in artificial intelligence. DARPA's 2016 Cyber Grand Challenge concentrated on autonomous systems and is a model other agencies can adopt, Clark said during a hearing of the House Oversight's Subcommittee on Information Technology "Every single agency has … problems it's going to encounter, and it has competitions that it can create to spur innovation, so it's not one single moonshot, it's a whole bunch of them," he told lawmakers during the hearing. "I think every part of government can contribute here." This is the third hearing this subcommittee has held on AI.