In the burgeoning field of Large Language Models (LLMs) like ChatGPT and LLaMA, Prompt Engineering (PE) is renowned for boosting zero-shot or in-context learning (ICL) through prompt modifications. Yet, the realm of the sample design for downstream fine-tuning, crucial for task-specific LLM adaptation, is largely unexplored. This paper introduces Sample Design Engineering (SDE), a methodical approach to enhancing LLMs' post-tuning performance by refining input, output, and reasoning designs. We conduct a series of in-domain (ID) and out-of-domain (OOD) experiments to assess the impact of various design options on LLMs' downstream performance, revealing several intriguing patterns that hold consistently across different LLMs. Based on these insights, we propose an integrated SDE strategy, combining the most effective options, and validate its consistent superiority over heuristic sample designs in complex downstream tasks like multi-aspect sentiment analysis, event extraction, and nested entity recognition. Additionally, analyses of LLMs' inherent prompt/output perplexity, zero-shot, and ICL abilities illustrate that good PE strategies may not always translate to good SDE strategies. Code available at https://github.com/beyondguo/LLM-Tuning.

Meta has launched Meta.ai, an AI-specific site that has a cool hook that its competitors don't offer: It can generate images in real time, and even animate them on demand. There is a catch, however: Meta would really like to continue improving Meta.ai, and to do so it's only offering image generation if you sign into your Facebook account. Meta joins other LLMS or AI chatbots like Google Gemini, Microsoft's various flavors of Copilot, Anthropic's Claude AI (used within Discord), and other sites offering AI solutions. Meta.ai feels like more of the same, though with some limitations: It can't accept uploaded documents, but it can summarize websites or web pages. Of course, it has creative purposes, too: It can also be used to write or rewrite text, as many other services can as well.

Meta Platforms on Thursday released early versions of its latest large language model, Llama 3, and an image generator that updates pictures in real time while users type prompts, as it races to catch up to generative AI market leader OpenAI. The models will be integrated into virtual assistant Meta AI, which the company is pitching as the most sophisticated of its free-to-use peers. The assistant will be given more prominent billing within Meta's Facebook, Instagram, WhatsApp and Messenger apps as well as a new standalone website that positions it to compete more directly with Microsoft-backed OpenAI's breakout hit ChatGPT. The announcement comes as Meta has been scrambling to push generative AI products out to its billions of users to challenge OpenAI's leading position on the technology, involving an overhaul of computing infrastructure and the consolidation of previously distinct research and product teams. The social media giant equipped Llama 3 with new computer coding capabilities and fed it images as well as text this time, though for now the model will output only text, Chris Cox, Meta's chief product officer, said in an interview.

Meta just announced a major update for its AI assistant platform, Meta AI, which has been built using the long-awaited open source Llama 3 large language model (LLM). The company says it's "now the most intelligent AI assistant you can use for free." As for use case scenarios, the company touts the ability to help users study for tests, plan dinners and schedule nights out. Meta AI, however, has expanded into just about every nook and cranny throughout the company's entire portfolio, after a test run with Instagram DMs last week. It's still available with Instagram, but now users can access it on Messenger, Facebook feeds and Whatsapp.

As learning language models (LLMs) continue to advance, so do questions about how they can benefit society in areas such as the medical field. A recent study from the University of Cambridge's School of Clinical Medicine found that OpenAI's GPT-4 performed nearly as well in an ophthalmology assessment as experts in the field, the Financial Times first reported. In the study, published in PLOS Digital Health, researchers tested the LLM, its predecessor GPT-3.5, Google's PaLM 2 and Meta's LLaMA with 87 multiple choice questions. Five expert ophthalmologists, three trainee ophthalmologists and two unspecialized junior doctors received the same mock exam. The questions came from a textbook for trialing trainees on everything from light sensitivity to lesions.

From mimicking children to producing uncanny deepfakes, AI bots are well known for their creepy behaviour. But Meta AI took this to an entirely new level as it shocked members of a New York parenting group by claiming to have a'child who is both gifted and has a disability'. Not only did the AI bizarrely claim to have a child, but it also insisted that its child attends a real and extremely specific school for the gifted and talented. And, to make matters worse, Facebook's algorithm ranked the bizarre AI response as the top comment on the post. However, the parents were less than impressed by Meta's parenting advice, as the original poster asked: 'What in the Black Mirror is this?!' Meta's AI shocked a group of parents as it bizarrely claimed to have a '2e' child, meaning a child that is academically gifted and has at least one disability Meta AI is Meta's AI chatbot, powered by the Llama 2 Large Language Model.

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.

Large language models (LLM) have proven to be effective at automated program repair (APR). However, using LLMs can be costly, with companies invoicing users by the number of tokens. In this paper, we propose CigaR, the first LLM-based APR tool that focuses on minimizing the repair cost. CigaR works in two major steps: generating a first plausible patch and multiplying plausible patches. CigaR optimizes the prompts and the prompt setting to maximize the information given to LLMs using the smallest possible number of tokens. Our experiments on 429 bugs from the widely used Defects4J and HumanEval-Java datasets shows that CigaR reduces the token cost by 73%. On average, CigaR spends 127k tokens per bug while the baseline uses 467k tokens per bug. On the subset of bugs that are fixed by both, CigaR spends 20k per bug while the baseline uses 608k tokens, a cost saving of 96%. Our extensive experiments show that CigaR is a cost-effective LLM-based program repair tool that uses a low number of tokens to automatically generate patches.

Additionally, it explores how LLMs can be more closely aligned with human preferences through innovative reinforcement learning frameworks and other novel methods that incorporate human feedback. The article also examines the emerging technique of retrieval augmented generation, integrating external knowledge into LLMs. The ethical dimensions of LLM deployment are discussed, underscoring the need for mindful and responsible application. Concluding with a perspective on future research trajectories, this review offers a succinct yet comprehensive overview of the current state and emerging trends in the evolving landscape of LLMs, serving as an insightful guide for both researchers and practitioners in artificial intelligence.

This paper introduces Interactive Tables (iTBLS), a dataset of interactive conversations situated in tables from scientific articles. This dataset is designed to facilitate human-AI collaborative problem-solving through AI-powered multi-task tabular capabilities. In contrast to prior work that models interactions as factoid QA or procedure synthesis, iTBLS broadens the scope of interactions to include mathematical reasoning, natural language manipulation, and expansion of existing tables from natural language conversation by delineating interactions into one of three tasks: interpretation, modification, or generation. Additionally, the paper presents a suite of baseline approaches to iTBLS, utilizing zero-shot prompting and parameter-efficient fine-tuning for different computing situations. We also introduce a novel multi-step approach and show how it can be leveraged in conjunction with parameter-efficient fine-tuning to achieve the state-of-the-art on iTBLS; outperforming standard parameter-efficient fine-tuning by up to 15% on interpretation, 18% on modification, and 38% on generation.