The strength of modern generative models lies in their ability to be controlled through text-based prompts. Typical "hard" prompts are made from interpretable words and tokens, and must be hand-crafted by humans. There are also "soft" prompts, which consist of continuous feature vectors. These can be discovered using powerful optimization methods, but they cannot be easily interpreted, re-used across models, or plugged into a text-based interface. We describe an approach to robustly optimize hard text prompts through efficient gradient-based optimization. Our approach automatically generates hard text-based prompts for both text-to-image and text-to-text applications. In the text-to-image setting, the method creates hard prompts for diffusion models, allowing API users to easily generate, discover, and mix and match image concepts without prior knowledge on how to prompt the model. In the text-to-text setting, we show that hard prompts can be automatically discovered that are effective in tuning LMs for classification.

Large language models (LLMs) have demonstrated solid zero-shot reasoning capabilities, which is reflected in their performance on the current test tasks. This calls for a more challenging benchmark requiring highly advanced reasoning ability to be solved. In this paper, we introduce such a benchmark, consisting of 191 long-form (1200 words on average) mystery narratives constructed as detective puzzles. Puzzles are sourced from the "5 Minute Mystery" platform and include a multiple-choice question for evaluation. Only 47% of humans solve a puzzle successfully on average, while the best human solvers achieve over 80% success rate. We show that GPT-3 models barely outperform random on this benchmark (with 28% accuracy) while state-of-the-art GPT-4 solves only 38% of puzzles. This indicates that there is still a significant gap in the deep reasoning abilities of LLMs and humans and highlights the need for further research in this area. Our work introduces a challenging benchmark for future studies on reasoning in language models and contributes to a better understanding of the limits of LLMs' abilities.

The window of what AI can't do seems to be contracting week by week. Machines can now write elegant prose and useful code, ace exams, conjure exquisite art, and predict how proteins will fold. Last summer I surveyed more than 550 AI researchers, and nearly half of them thought that, if built, high-level machine intelligence would lead to impacts that had at least a 10% chance of being "extremely bad (e.g. On May 30, hundreds of AI scientists, along with the CEOs of top AI labs like OpenAI, DeepMind and Anthropic, signed a statement urging caution on AI: "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war." The simplest argument is that progress in AI could lead to the creation of superhumanly-smart artificial "people" with goals that conflict with humanity's interests--and the ability to pursue them autonomously.

With the rise of AI language models and tools like ChatGPT and Bard, we've heard warnings from people involved, like Elon Musk, about the risks posed by AI. Now, a group of high-profile industry leaders has issued a one-sentence statement: "Mitigating the risk of extinction from AI should be a global priority alongside other societal-scale risks such as pandemics and nuclear war." It was posted to the Center for AI Safety, an organization with the mission "to reduce societal-scale risks from artificial intelligence," according to its website. Signatories include OpenAI chief executive Sam Altman and Google DeepMind head Demis Hassabis. Turing Award-winning researchers Geoffrey Hinton and Yoshua Bengio, the godfathers of modern AI, also put their names to it.

Computer scientist Pascale Fung can imagine a rosy future in which polyglot AI helpers like ChatGPT bridge language barriers. In that world, Indonesian store owners fluent only in local dialects might reach new shoppers by listing their products online in English. "It can open opportunities," Fung says--then pauses. She's spotted the bias in her vision of a more interconnected future: The AI-aided shopping would be one-sided, because few Americans would bother to use AI translation to help research products advertised in Indonesian. "Americans are not incentivized to learn another language," she says.

Sen. Josh Hawley sat down with Fox News Digital for a wide-ranging interview about his new book, "Manhood: The Masculine Virtues America Needs." Senator Josh Hawley, R-Mo., is very concerned about the power of Artificial Intelligence to "manipulate Americans and the "facts" they are given from the technology on a daily basis. "I'm worried about AI's power to manipulate our attention, to manipulate our opinions and to manipulate the information that we're given," he told Fox News Digital in a recent in-person interview. The Missouri senator, the ranking member on the Senate Judiciary Subcommittee on Privacy, Technology and the Law, continued, "Already you can see these generative AI systems – these large language models – that are trained on all the information on the internet." AI WILL BE THE POLITICAL LEFT'S'SINGLE GREATEST WEAPON' AGAINST RELIGIOUS FAITH AND TRUTH, SAYS EXPERT He added, "You can train them on your own.

Taipei, Taiwan – Artificial intelligence poses a "risk of extinction" that calls for global action, leading computer scientists and technologists have warned. "Mitigating the risk of extinction from AI should be a global priority alongside other societal scale risks such as pandemics and nuclear war," a group of AI experts and other high-profile figures said in a brief statement released by the Center for AI Safety, a San Francisco-based research and advocacy group, on Tuesday. The signatories include technology experts such as Sam Altman, chief executive of OpenAI, Geoffrey Hinton, known as the "godfather of AI", and Audrey Tang, Taiwan's digital minister, as well as other notable figures including the neuroscientist Sam Harris and the musician Grimes. The warning follows an open letter signed by Elon Musk and other high-profile figures in March that called for a six-month pause on the development of AI more advanced than OpenAI's GPT-4. "Powerful AI systems should be developed only once we are confident that their effects will be positive and their risks will be manageable," the letter said.

For prediction and real-time control tasks, machine-learning (ML)-based digital twins are frequently employed. However, while these models are typically accurate, they are custom-designed for individual systems, making system-to-system (S2S) transferability difficult. This occurs even when substantial similarities exist in the process dynamics across different chemical systems. To address this challenge, we developed a novel time-series-transformer (TST) framework that exploits the powerful transfer learning capabilities inherent in transformer algorithms. This was demonstrated using readily available process data obtained from different crystallizers operating under various operational scenarios. Using this extensive dataset, we trained a TST model (CrystalGPT) to exhibit remarkable S2S transferability not only across all pre-established systems, but also to an unencountered system. CrystalGPT achieved a cumulative error across all systems, which is eight times superior to that of existing ML models. Additionally, we coupled CrystalGPT with a predictive controller to reduce the variance in setpoint tracking to just 1%.

A central tenet of Federated learning (FL), which trains models without centralizing user data, is privacy. However, previous work has shown that the gradient updates used in FL can leak user information. While the most industrial uses of FL are for text applications (e.g. keystroke prediction), nearly all attacks on FL privacy have focused on simple image classifiers. We propose a novel attack that reveals private user text by deploying malicious parameter vectors, and which succeeds even with mini-batches, multiple users, and long sequences. Unlike previous attacks on FL, the attack exploits characteristics of both the Transformer architecture and the token embedding, separately extracting tokens and positional embeddings to retrieve high-fidelity text. This work suggests that FL on text, which has historically been resistant to privacy attacks, is far more vulnerable than previously thought.

Temporal reasoning is the task of predicting temporal relations of event pairs. While temporal reasoning models can perform reasonably well on in-domain benchmarks, we have little idea of these systems' generalizability due to existing datasets' limitations. In this work, we introduce a novel task named TODAY that bridges this gap with temporal differential analysis, which as the name suggests, evaluates whether systems can correctly understand the effect of incremental changes. Specifically, TODAY introduces slight contextual changes for given event pairs, and systems are asked to tell how this subtle contextual change would affect relevant temporal relation distributions. To facilitate learning, TODAY also annotates human explanations. We show that existing models, including GPT-3.5, drop to random guessing on TODAY, suggesting that they heavily rely on spurious information rather than proper reasoning for temporal predictions. On the other hand, we show that TODAY's supervision style and explanation annotations can be used in joint learning, encouraging models to use more appropriate signals during training and thus outperform across several benchmarks. TODAY can also be used to train models to solicit incidental supervision from noisy sources such as GPT-3.5, thus moving us more toward the goal of generic temporal reasoning systems.