Developing safe and useful general-purpose AI systems will require us to make progress on scalable oversight: the problem of supervising systems that potentially outperform us on most skills relevant to the task at hand. Empirical work on this problem is not straightforward, since we do not yet have systems that broadly exceed our abilities. This paper discusses one of the major ways we think about this problem, with a focus on ways it can be studied empirically. We first present an experimental design centered on tasks for which human specialists succeed but unaided humans and current general AI systems fail. We then present a proof-of-concept experiment meant to demonstrate a key feature of this experimental design and show its viability with two question-answering tasks: MMLU and time-limited QuALITY. On these tasks, we find that human participants who interact with an unreliable large-language-model dialog assistant through chat -- a trivial baseline strategy for scalable oversight -- substantially outperform both the model alone and their own unaided performance. These results are an encouraging sign that scalable oversight will be tractable to study with present models and bolster recent findings that large language models can productively assist humans with difficult tasks.

Associative memory plays an important role in human intelligence and its mechanisms have been linked to attention in machine learning. While the machine learning community's interest in associative memories has recently been rekindled, most work has focused on memory recall ($read$) over memory learning ($write$). In this paper, we present BayesPCN, a hierarchical associative memory capable of performing continual one-shot memory writes without meta-learning. Moreover, BayesPCN is able to gradually forget past observations ($forget$) to free its memory. Experiments show that BayesPCN can recall corrupted i.i.d. high-dimensional data observed hundreds to a thousand ``timesteps'' ago without a large drop in recall ability compared to the state-of-the-art offline-learned parametric memory models.

Despite being well researched in many languages, information retrieval has still not received much attention from the Vietnamese research community. This is especially true for the case of legal documents, which are hard to process. This study proposes a new approach for information retrieval for Vietnamese legal documents using sentence-transformer. Besides, various experiments are conducted to make comparisons between different transformer models, ranking scores, syllable-level, and word-level training. The experiment results show that the proposed model outperforms models used in current research on information retrieval for Vietnamese documents.

Detroit, Michigan, Nov. 02, 2022 (GLOBE NEWSWIRE) -- Artificial Intelligence Technology Solutions, Inc., (the Company) (OTCPK:AITX), has announced that on October 28, 2022 it issued a $4 million note to its largest single investor thereby securing a loan that matures in 4 years, bears interest at 15% per annum, has an original issue discount of $500,000, provides cash proceeds to the Company of $3.5 million, and includes warrants to acquire additional preferred equity shares (the "Fundraise"). The net effect of the Fundraise does not materially affect the Company's common stock shareholders or common shareholders' equity percentage since Steve Reinharz, AITX Founder and CEO, has effectively reduced his stake by approximately 20% (from fully diluted ownership of 65% to 54%) to achieve the funding without any further dilution to common shareholders. Steve Reinharz commented, "My commitment has always been to make AITX along with its RAD subsidiaries the dominant player in the evolving #proptech industry, which we feel we helped write the book on. This funding is crucial to the Company and will allow us to continue to grow while adding potential value to all stakeholders." Reinharz continued, "Certainly, I don't love reducing my overall stake, but the way this deal issues preferred shares and doesn't affect common shareholders, is a significant demonstration of my commitment to the Company's mission and to our shareholders."

But just how smart are these large language models? On the last day of the conference, I interviewed legendary linguist Noam Chomsky, now 93 years old, and Gary Marcus, an emeritus professor of cognitive science at New York University who has spent much of the past decade highlighting the limits of deep learning. Both were distinctly unimpressed with today's cutting edge A.I. Chomsky's big disappointment is that these large language models don't tell us anything at all about how the human brain works. Chomsky has devoted much of his life to advancing the theory that there is a universal grammar, or at least a set of structural concepts, that underpin all human languages, and that this grammar is somehow hard-wired into the brain. Chomsky thinks this explains why human infants can master language so easily--whereas today's computer systems need to be fed what Chomsky rightly calls "astronomical amounts of data" and even then still don't actually understand language at all.

If you follow the media stories about AI, you will see two schools of thought. One school is utopian, proclaiming the amazing power of AI, from predicting quantum electron paths to driving a race car like a champion. The other school is dystopian, scaring us with crisis-ridden stories that range from how AI could bring about the end of privacy to self-driving cars that almost immediately crash. One school of thought is outraged by imperfection, while the other lives in denial. But neither extreme view accurately represents our imperfect world. As Stephen Hawking said, "One of the basic rules of the universe is that nothing is perfect.

While GitHub continues to bolster its Copilot service with new features, the software has also been targeted with a proposed class-action lawsuit. If the lawsuit is granted class-action status, it could upend the defense that such data collection is covered in the US by fair use doctrine, potentially affecting not only the legality of Copilot but also a whole range of generative AI models.

This week, Microsoft President and vice chair Brad Smith is heading to Egypt for the United Nation's annual climate conference with a mission: show the world that the tech giant is "consistent and committed" in its climate goals, as well as communicate the "vital role" that the tech industry as a whole has to play in battling the climate crisis. The Microsoft leader has been busy in recent months since the departure of chief environmental officer Lucas Joppa, stepping in to lead the company's climate initiatives (something Smith has always been intimately involved with, as Joppa's boss prior to his departure). Last week at the Web Summit tech conference, he spoke about the urgency of the workforce transformation the world needs to reach net zero, as well as the current skills gap. "The key to the future is going to be a new generation of people with a new generation of technology coming from a new generation of companies," he said, highlighting the work of startups like the India-based SEEDS, which is using satellite data and AI to identify homes that would be most susceptible to extreme heat, then helping them adapt. Using AI and data to help the Global South adapt to climate change is one of Microsoft's main focuses going into the COP27 climate talks.

Several recent studies [8, 41, 55, 67, 75] propose various learning strategies for AI models to be well-calibrated across all protected subgroups, while others focus on collecting responsible datasets [57, 82, 124] to make sure evaluations of AI models are accurate and algorithmic bias can be measured while promoting data privacy. There has been much criticism regarding the design choice of the publicly used datasets, such as for ImageNet [36, 38, 56, 70]. Discussions are mostly focused on concerns around collecting sensitive data about people without their consent. Casual Conversations v1 [57] was one of the first benchmarks that was designed with permission from participants. However, that dataset has several limitations: samples were collected only in the US, the gender label is limited to three options, and only age and gender labels are self-provided with the permission of the participants.

Privacy-preserving machine learning in data-sharing processes is an ever-critical task that enables collaborative training of Machine Learning (ML) models without the need to share the original data sources. It is especially relevant when an organization must assure that sensitive data remains private throughout the whole ML pipeline, i.e., training and inference phases. This paper presents an innovative framework that uses Representation Learning via autoencoders to generate privacy-preserving embedded data. Thus, organizations can share the data representation to increase machine learning models' performance in scenarios with more than one data source for a shared predictive downstream task.