Pretrained Language Models (PLMs) harbor inherent social biases that can result in harmful real-world implications. Such social biases are measured through the probability values that PLMs output for different social groups and attributes appearing in a set of test sentences. However, bias testing is currently cumbersome since the test sentences are generated either from a limited set of manual templates or need expensive crowd-sourcing. We instead propose using ChatGPT for the controllable generation of test sentences, given any arbitrary user-specified combination of social groups and attributes appearing in the test sentences. When compared to template-based methods, our approach using ChatGPT for test sentence generation is superior in detecting social bias, especially in challenging settings such as intersectional biases. We present an open-source comprehensive bias testing framework (BiasTestGPT), hosted on HuggingFace, that can be plugged into any open-source PLM for bias testing. User testing with domain experts from various fields has shown their interest in being able to test modern AI for social biases. Our tool has significantly improved their awareness of such biases in PLMs, proving to be learnable and user-friendly. We thus enable seamless open-ended social bias testing of PLMs by domain experts through an automatic large-scale generation of diverse test sentences for any combination of social categories and attributes.

The ChatGPT developer's new board of directors and its briefly fired but now-restored CEO, Sam Altman, said last week that they're trying to fix the unusual corporate structure that allowed four board members to trigger a near-death experience for the company. The startup was founded in 2015 as a nonprofit, but it develops AI inside a capped-profit subsidiary answerable to the nonprofit's board, which is charged with ensuring that the technology is "broadly beneficial" to humanity. To stabilize this unusual structure, OpenAI could take pointers from longer-lived companies with a similar arrangement--including introducing a second board to help balance its founding mission with its for-profit pursuit of returns for investors. OpenAI deferred comment for this story to new board chair Bret Taylor. The veteran tech executive told WIRED in a statement that the board is focused on overseeing an independent review of the recent crisis and enhancing governance.

Facebook parent Meta and IBM on Tuesday launched a new group called the AI Alliance advocating for an "open-science" approach to AI development that puts them at odds with rivals Google, Microsoft and ChatGPT-maker OpenAI. These two diverging camps – the open and the closed – disagree about whether to build AI in a way that makes the underlying technology widely accessible. Safety is at the heart of the debate, but so is who gets to profit from AI's advances. Open advocates favor an approach that is "not proprietary and closed", said Darío Gil, a senior vice-president at IBM who directs its research division. "So it's not like a thing that is locked in a barrel and no one knows what they are."

For the first birthday of Microsoft Copilot, Microsoft laid out a roadmap of additional features that it plans to add, including a more sophisticated language model and better ways of generating AI art. Microsoft plans to add these things to Copilot, which is now the company's catchall term for its large language model chatbots on the web, on Windows, and in Microsoft 365. Of particular interest is the imminent addition of GPT-4 Turbo, notable for two things: an information cutoff of April 2023 as well as the ability to handle 128k of input (or about 300 pages of text.) The feature is currently in testing, Microsoft said. Still, there is a cutoff -- and to solve that problem, Microsoft is launching Deep Search.

Microsoft just announced its Copilot AI chatbot is integrating with OpenAI's latest model, GPT-4 Turbo, and the image generator DALL-E 3, among other upgrades. This should drastically improve the overall functionality of the service, just in time for its one-year anniversary/birthday. Wait, do AI chatbots have birthdays? GPT-4 Turbo integration will allow Copilot users to tackle complex tasks that would cause previous iterations of the software to sputter into madness. The last generation allowed for just 50 pages of text as a data input, while GPT-4 Turbo accepts up to 300 pages. The integration is currently being tested by select users, with wider availability in the next few weeks.

When the board of OpenAI suddenly fired the company's CEO last month, it sparked speculation that board members were rattled by the breakneck pace of progress in artificial intelligence and the possible risks of seeking to commercialize the technology too quickly. Robust Intelligence, a startup founded in 2020 to develop ways to protect AI systems from attack, says that some existing risks need more attention. Working with researchers from Yale University, Robust Intelligence has developed a systematic way to probe large language models (LLMs), including OpenAI's prized GPT-4 asset, using "adversarial" AI models to discover "jailbreak" prompts that cause the language models to misbehave. While the drama at OpenAI was unfolding, the researchers warned OpenAI of the vulnerability. They say they have yet to receive a response.

AI development and concerns about its safety continue to grow at a rapid pace with little regulation in place. The latest industry-based solution to this comes courtesy of IBM and Meta, which have announced the creation of the AI Alliance. Its mission centers on "fostering an open community and enabling developers and researchers to accelerate responsible innovation in AI while ensuring scientific rigor, trust, safety, security, diversity and economic competitiveness." Part of this work will involve efforts to expand the number of open-source AI models -- ones with public source code -- which runs counter to the private models of companies like OpenAI and Google. Open-sourcing is a key pillar of the AI Alliance.

The recent OpenAI saga, in which Microsoft exerted its quiet but firm dominance over the "capped profit" entity, provides a powerful demonstration of what we've been analyzing for the last half-decade. To wit: those with the money make the rules. And right now, they're engaged in a race to the bottom, releasing systems before they're ready in an attempt to retain their dominant position. Relying on a few unaccountable corporate actors for core infrastructure is a problem for democracy, culture, and individual and collective agency. Without significant intervention, the AI market will only end up rewarding and entrenching the very same companies that reaped the profits of the invasive surveillance business model that has powered the commercial internet, often at the expense of the public. The Cambridge Analytica scandal was just one among many that exposed this seedy reality.

Conversational question-answering (CQA) systems aim to create interactive search systems that effectively retrieve information by interacting with users. To replicate human-to-human conversations, existing work uses human annotators to play the roles of the questioner (student) and the answerer (teacher). Despite its effectiveness, challenges exist as human annotation is time-consuming, inconsistent, and not scalable. To address this issue and investigate the applicability of large language models (LLMs) in CQA simulation, we propose a simulation framework that employs zero-shot learner LLMs for simulating teacher-student interactions. Our framework involves two LLMs interacting on a specific topic, with the first LLM acting as a student, generating questions to explore a given search topic. The second LLM plays the role of a teacher by answering questions and is equipped with additional information, including a text on the given topic. We implement both the student and teacher by zero-shot prompting the GPT-4 model. To assess the effectiveness of LLMs in simulating CQA interactions and understand the disparities between LLM- and human-generated conversations, we evaluate the simulated data from various perspectives. We begin by evaluating the teacher's performance through both automatic and human assessment. Next, we evaluate the performance of the student, analyzing and comparing the disparities between questions generated by the LLM and those generated by humans. Furthermore, we conduct extensive analyses to thoroughly examine the LLM performance by benchmarking state-of-the-art reading comprehension models on both datasets. Our results reveal that the teacher LLM generates lengthier answers that tend to be more accurate and complete. The student LLM generates more diverse questions, covering more aspects of a given topic.

Chain-of-Thought (CoT) guides large language models (LLMs) to reason step-by-step, and can motivate their logical reasoning ability. While effective for logical tasks, CoT is not conducive to creative problem-solving which often requires out-of-box thoughts and is crucial for innovation advancements. In this paper, we explore the Leap-of-Thought (LoT) abilities within LLMs -- a non-sequential, creative paradigm involving strong associations and knowledge leaps. To this end, we study LLMs on the popular Oogiri game which needs participants to have good creativity and strong associative thinking for responding unexpectedly and humorously to the given image, text, or both, and thus is suitable for LoT study. Then to investigate LLMs' LoT ability in the Oogiri game, we first build a multimodal and multilingual Oogiri-GO dataset which contains over 130,000 samples from the Oogiri game, and observe the insufficient LoT ability or failures of most existing LLMs on the Oogiri game. Accordingly, we introduce a creative Leap-of-Thought (CLoT) paradigm to improve LLM's LoT ability. CLoT first formulates the Oogiri-GO dataset into LoT-oriented instruction tuning data to train pretrained LLM for achieving certain LoT humor generation and discrimination abilities. Then CLoT designs an explorative self-refinement that encourages the LLM to generate more creative LoT data via exploring parallels between seemingly unrelated concepts and selects high-quality data to train itself for self-refinement. CLoT not only excels in humor generation in the Oogiri game but also boosts creative abilities in various tasks like cloud guessing game and divergent association task. These findings advance our understanding and offer a pathway to improve LLMs' creative capacities for innovative applications across domains. The dataset, code, and models will be released online. https://zhongshsh.github.io/CLoT/.