By submitting this content request, I have legitimate interest in the content and agree that Technology Innovation Institute, their partners, and the creators of any other content I have selected may contact me regarding news, products, and services that may be of interest to me. By submitting this content request, I have legitimate interest in the content and agree that Technology Innovation Institute, their partners, and the creators of any other content I have selected may contact me regarding news, products, and services that may be of interest to me. I agree to the IEEE Privacy Policy Are you an IEEE member?

As more companies rush to implement AI solutions and software, a growing number of experts are warning that it could result in an explosion of'fake news' and misinformation. The United States and Italy agreed on Wednesday to coordinate efforts to counter the spread of misinformation and fake news articles by foreign governments. U.S. Secretary of State Antony Blinken and Italian Foreign Minister Antonio Tajani agreed on the new pact during a meeting on the sidelines of a three-day meeting of Group of Seven (G7) foreign ministers on the island of Capri. The U.S. last year released an intelligence assessment sent to more than 100 countries that accused Moscow of using spies, social media and Russian state-run media to erode public faith in the integrity of democratic elections. Last week, Belgium said its prosecutors were probing alleged Russian attempts to influence an upcoming European Parliamentary election.

For a glimpse of where artificial intelligence is headed in election campaigns, look to India, the world's largest democracy, as it starts heading to the polls Friday. An AI-generated version of Prime Minister Narendra Modi that has been shared on WhatsApp shows the possibilities for hyperpersonalized outreach in a country with nearly 1 billion voters. In the video -- a demo clip the source of which is unclear -- Modi's avatar addresses a series of voters directly, by name. However, it is not perfect. Modi appears to wear two different pairs of glasses, and some parts of the video are pixelated.

Israel Defense Forces spokesperson Rear Adm. Daniel Hagari discusses Iran's attack on Israel, saying the attacks proved that Iran seeks to "escalate the region." JERUSALEM -- Some of Israel's most advanced military technology was on display over the weekend when its multi-level aerial defense array led the way in striking down an estimated 99% of the more than 350 drones, rockets and missiles that were fired by Iran in an unprecedented attack on the Jewish state. From the Iron Dome, which in its latest format uses artificial intelligence (AI) to improve accuracy when shooting short-range surface-to-surface rockets, to David's Sling, which intercepts short- to medium-range and medium- to long-range surface-to-surface missiles, to the Arrow 2 and 3 systems, which is used for longer-range ballistic and cruise missiles, as well as AI-driven aircraft and other technology, Israel's defensive operation proved it was far superior to the offensive capabilities of the Islamic Republic. In a press briefing following the attack, Israel Defense Forces spokesperson Rear Adm. Daniel Hagari hailed Israel's defensive operation, which was carried out together with partners from U.S. Central Command (CENTCOM), as a "very significant strategic achievement." He said it demonstrated the "exceptional professionalism" of Israel's Aerial Defense Array and the "defensive abilities of the air force as well as the army's military and technological superiority."

When the system detects small cracks in road surfaces, it promptly seals them. Over 112,000 Americans overdosed in 2023. Over 70,000 overdosed on fentanyl, a particularly dangerous synthetic opioid. While it's easy to focus on numbers, we can't forget that these aren't statistics. They leave behind parents, brothers, sisters and children.

Russia is working to undermine U.S. support for Ukraine ahead of the 2024 election, while Chinese influence campaigns aim to divide Americans and undercut their faith in democracy, according to new research from Microsoft. But both countries' persistent efforts to shape American politics through propaganda and cyberattacks have been less pronounced during this year's election cycle than they were in 2016 and 2020, the company said in a report published Wednesday. It also found that widespread fears of voters being influenced by deepfake videos made with artificial intelligence haven't yet happened, while simpler efforts have been more persuasive. "China has always adhered to non-interference in other countries' internal affairs," said Liu Pengyu, a spokesperson for the Chinese Embassy in Washington. "Speculating or accusing China of using social media to interfere in the U.S. politics is completely groundless and malicious speculation."

Hallucination continues to be one of the most critical challenges in the institutional adoption journey of Large Language Models (LLMs). In this context, an overwhelming number of studies have focused on analyzing the post-generation phase - refining outputs via feedback, analyzing logit output values, or deriving clues via the outputs' artifacts. We propose HalluciBot, a model that predicts the probability of hallucination $\textbf{before generation}$, for any query imposed to an LLM. In essence, HalluciBot does not invoke any generation during inference. To derive empirical evidence for HalluciBot, we employ a Multi-Agent Monte Carlo Simulation using a Query Perturbator to craft $n$ variations per query at train time. The construction of our Query Perturbator is motivated by our introduction of a new definition of hallucination - $\textit{truthful hallucination}$. Our training methodology generated 2,219,022 estimates for a training corpus of 369,837 queries, spanning 13 diverse datasets and 3 question-answering scenarios. HalluciBot predicts both binary and multi-class probabilities of hallucination, enabling a means to judge the query's quality with regards to its propensity to hallucinate. Therefore, HalluciBot paves the way to revise or cancel a query before generation and the ensuing computational waste. Moreover, it provides a lucid means to measure user accountability for hallucinatory queries.

Language models (LMs) trained on vast quantities of text data can acquire sophisticated skills such as generating summaries, answering questions or generating code. However, they also manifest behaviors that violate human preferences, e.g., they can generate offensive content, falsehoods or perpetuate social biases. In this thesis, I explore several approaches to aligning LMs with human preferences. First, I argue that aligning LMs can be seen as Bayesian inference: conditioning a prior (base, pretrained LM) on evidence about human preferences (Chapter 2). Conditioning on human preferences can be implemented in numerous ways. In Chapter 3, I investigate the relation between two approaches to finetuning pretrained LMs using feedback given by a scoring function: reinforcement learning from human feedback (RLHF) and distribution matching. I show that RLHF can be seen as a special case of distribution matching but distributional matching is strictly more general. In chapter 4, I show how to extend the distribution matching to conditional language models. Finally, in chapter 5 I explore a different root: conditioning an LM on human preferences already during pretraining. I show that involving human feedback from the very start tends to be more effective than using it only during supervised finetuning. Overall, these results highlight the room for alignment techniques different from and complementary to RLHF.

Can Large Language Models substitute humans in making important decisions? Recent research has unveiled the potential of LLMs to role-play assigned personas, mimicking their knowledge and linguistic habits. However, imitative decision-making requires a more nuanced understanding of personas. In this paper, we benchmark the ability of LLMs in persona-driven decision-making. Specifically, we investigate whether LLMs can predict characters' decisions provided with the preceding stories in high-quality novels. Leveraging character analyses written by literary experts, we construct a dataset LIFECHOICE comprising 1,401 character decision points from 395 books. Then, we conduct comprehensive experiments on LIFECHOICE, with various LLMs and methods for LLM role-playing. The results demonstrate that state-of-the-art LLMs exhibit promising capabilities in this task, yet there is substantial room for improvement. Hence, we further propose the CHARMAP method, which achieves a 6.01% increase in accuracy via persona-based memory retrieval. We will make our datasets and code publicly available.

Large Language Models (LLMs) exhibit impressive zero/few-shot inference and generation quality for high-resource languages (HRLs). A few of them have been trained on low-resource languages (LRLs) and give decent performance. Owing to the prohibitive costs of training LLMs, they are usually used as a network service, with the client charged by the count of input and output tokens. The number of tokens strongly depends on the script and language, as well as the LLM's subword vocabulary. We show that LRLs are at a pricing disadvantage, because the well-known LLMs produce more tokens for LRLs than HRLs. This is because most currently popular LLMs are optimized for HRL vocabularies. Our objective is to level the playing field: reduce the cost of processing LRLs in contemporary LLMs while ensuring that predictive and generative qualities are not compromised. As means to reduce the number of tokens processed by the LLM, we consider code-mixing, translation, and transliteration of LRLs to HRLs. We perform an extensive study using the IndicXTREME classification and six generative tasks dataset, covering 15 Indic and 3 other languages, while using GPT-4 (one of the costliest LLM services released so far) as a commercial LLM. We observe and analyze interesting patterns involving token count, cost, and quality across a multitude of languages and tasks. We show that choosing the best policy to interact with the LLM can reduce cost by 90% while giving better or comparable performance compared to communicating with the LLM in the original LRL.