GPT-3 and Humor Generation; Reasoning and Acting in LMs; Claude-2 vs. GPT-4; Google Bard Access; Expert Advice; and More

Editor's Paper Recommendations

Prompt to GPT-3: Step-by-Step Thinking Instructions for Humor Generation: Artificial intelligence has made significant progress in natural language processing, with models like GPT-3 demonstrating impressive capabilities. However, these models still need to be improved in complex tasks that require understanding the user, such as mastering human comedy writing strategies. This paper explores humor generation using GPT-3 by modeling human comedy writing theory and leveraging step-by-step thinking instructions. In addition, we explore the role of cognitive distance in creating humor.

Investigating Pre-trained Language Models on Cross-Domain Datasets, a Step Closer to General AI: In this paper, the researchers investigate the generalization ability of pre-trained language models across different non-language tasks. They evaluate four pre-trained models (T5, BART, BERT, and GPT-2) on tasks from various domains, including computer vision, hierarchical reasoning, and protein fold prediction. The pre-trained models consistently outperform transformers trained from scratch by a significant margin, demonstrating the effectiveness of pre-training on language tasks. The study also shows that reducing the number of parameters in pre-trained models has minimal impact on performance, and using pre-trained embeddings for the input layer is necessary for achieving desired results. The findings suggest that pre-trained language models contribute to acquiring general knowledge and advancing toward the goal of general AI.

Kosmos-2: Grounding Multimodal Large Language Models to the World: We introduce Kosmos-2, a Multimodal Large Language Model (MLLM), enabling new capabilities of perceiving object descriptions (e.g., bounding boxes) and grounding text to the visual world. Specifically, we represent refer expressions as links in Markdown, i.e., ``[text span](bounding boxes)'', where object descriptions are sequences of location tokens. We construct large-scale data of grounded image-text pairs (called GrIT) to train the model with multimodal corpora. In addition to the existing capabilities of MLLMs (e.g., perceiving general modalities, following instructions, and performing in-context learning), Kosmos-2 integrates the grounding capability into downstream applications. We evaluate Kosmos-2 on a wide range of tasks, including (i) multimodal groundings, such as referring expression comprehension, and phrase grounding, (ii) multimodal referring, such as referring expression generation, (iii) perception-language tasks; and (iv) language understanding and generation. This work lays the foundation for the development of Embodiment AI. It sheds light on the convergence of language, multimodal perception, action, and world modeling, a key step toward artificial general intelligence. Data, demo, and pre-trained models are available at?this https URL.

ReAct: Synergizing Reasoning and Acting in Language Models: While large language models (LLMs) have demonstrated impressive capabilities across tasks in language understanding and interactive decision-making, their abilities for reasoning (e.g., chain-of-thought prompting) and acting (e.g., action plan generation) have primarily been studied as separate topics. In this paper, we explore the use of LLMs to generate both reasoning traces and task-specific actions in an interleaved manner, allowing for greater synergy between the two: reasoning traces help the model induce, track, and update action plans as well as handle exceptions, while actions allow it to interface with external sources, such as knowledge bases or environments, to gather additional information. We apply our approach, named ReAct, to a diverse set of language and decision-making tasks and demonstrate its effectiveness over state-of-the-art baselines and improved human interpretability and trustworthiness over methods without reasoning or acting components. Concretely, on question answering (HotpotQA) and fact verification (Fever), ReAct overcomes issues of hallucination and error propagation prevalent in chain-of-thought reasoning by interacting with a simple Wikipedia API and generates human-like task-solving trajectories that are more interpretable than baselines without reasoning traces. On two interactive decision-making benchmarks (ALFWorld and WebShop), ReAct outperforms imitation and reinforcement learning methods by an absolute success rate of 34% and 10%, respectively, while being prompted with only one or two in-context examples. Project site with code:?this https URL

Industry Insights

• Claude-2 vs. GPT-4
• Why OpenAI removed Web Browsing from ChatGPT
• How to Get Access to Google Bard and Its API in July 2023
• Meta's new AI model aims to break the dominance of OpenAI and GPT-4

Webinar on ChatGPT Plugin Store

The ChatGPT Plugin Store is an exciting platform that offers developers a wide array of innovative and powerful plugins to enhance the capabilities of ChatGPT, including artificial intelligence and machine learning. These plugins serve as modular extensions that enable ChatGPT to perform specific tasks, making it a versatile tool for various domains and applications.

Webinar Registration Link

Weekly Concept Breakdown

Large Language Model

Discover the fascinating world of large language models in the article "A High-level Overview of Large Language Models." Delve into their advanced text generation capabilities, understand the key components of these models, such as token embeddings and decoder-only Transformers, and explore their training process. From input representation to next-word selection, scaling laws, and fine-tuning methods, this concise overview provides insights into the inner workings of these powerful models and their potential for transforming natural language processing.

Growth Zone

• Use Gratitude to Counter Stress and Uncertainty

Expert Advice