Transformer-based large language models (LLMs) provide a powerful foundation for natural language tasks in large-scale customer-facing applications. However, studies that explore their vulnerabilities emerging from malicious user interaction are scarce. By proposing PromptInject, a prosaic alignment framework for mask-based iterative adversarial prompt composition, we examine how GPT-3, the most widely deployed language model in production, can be easily misaligned by simple handcrafted inputs. In particular, we investigate two types of attacks -- goal hijacking and prompt leaking -- and demonstrate that even low-aptitude, but sufficiently ill-intentioned agents, can easily exploit GPT-3's stochastic nature, creating long-tail risks. The code for PromptInject is available at https://github.com/agencyenterprise/PromptInject.

Large-scale transformer models have become the de-facto architectures for various machine learning applications, e.g., CV and NLP. However, those large models also introduce prohibitive training costs. To mitigate this issue, we propose a novel random and layerwise token dropping method (random-LTD), which skips the computation of a subset of the input tokens at all middle layers. Particularly, random-LTD achieves considerable speedups and comparable accuracy as the standard training baseline. Compared to other token dropping methods, random-LTD does not require (1) any importance score-based metrics, (2) any special token treatment (e.g., [CLS]), and (3) many layers in full sequence length training except the first and the last layers. Besides, a new LayerToken learning rate schedule is proposed for pretraining problems that resolve the heavy tuning requirement for our proposed training mechanism. Finally, we demonstrate that random-LTD can be applied to broader applications, including GPT and BERT pretraining as well as ViT and GPT finetuning tasks. Our results show that random-LTD can save about 33.3% theoretical compute cost and 25.6% wall-clock training time while achieving similar zero-shot evaluations on GPT-31.3B as compared to baseline.

"what I find is that it's a very bizarre mixture of ideas that are solid and good with ideas that are crazy. It's as if you took a lot of very good food and some dog excrement and blended it all up so that you can't possibly figure out what's good or bad." I can't wait to see the fawning New York Times story tomorrow morning. But…wait…well, um, how do I put this politely? Just like every other large language model I have seen.

GPT-4 is possibly the most anticipated AI model in history. In 2020, GPT-3 surprised everyone with a huge performance leap from GPT-2 and set unprecedented expectations for its successor. But for two years OpenAI has been super shy about GPT-4--letting out info in dribs and drabs and remaining silent for the most part. People have been talking these months. What I've heard from several sources: GPT-4 is almost ready and will be released (hopefully) sometime December-February.

But the real problem is that, especially without extensive AI expertise, smaller platforms can only do so much. The agenda so far is being set by fast-moving AI startups like OpenAI and Stability, as well as tech giants like Google. Beyond simply banning AI-generated work, there's no easy way to navigate the system without touching what's become a third rail to many artists. "This is not something that DeviantArt can fix on our own," admits Gurwicz. "Until there's proper regulation in place, it does require these AI models and platforms to go beyond just what is legally required and think about, ethically, what's right and what's fair."

Webサイト制作において、ターゲットに合わせた訴求文を作ったり、掲載場所に応じて文章の長さを変えたりする「テキスト作成」は、簡単なようで時間や労力がかかるものです。 そんな中、注目されているのが文章生成AIモデル「GPT...

As machine learning permeates more industries and models become more expensive and time consuming to train, the need for efficient automated hyperparameter optimization (HPO) has never been more pressing. Multi-step planning based approaches to hyperparameter optimization promise improved efficiency over myopic alternatives by more effectively balancing out exploration and exploitation. However, the potential of these approaches has not been fully realized due to their technical complexity and computational intensity. In this work, we leverage recent advances in Transformer-based, natural-language-interfaced hyperparameter optimization to circumvent these barriers. We build on top of the recently proposed OptFormer which casts both hyperparameter suggestion and target function approximation as autoregressive generation thus making planning via rollouts simple and efficient. We conduct extensive exploration of different strategies for performing multi-step planning on top of the OptFormer model to highlight its potential for use in constructing non-myopic HPO strategies.

Information overload is a major obstacle to scientific progress. The explosive growth in scientific literature and data has made it ever harder to discover useful insights in a large mass of information. Today scientific knowledge is accessed through search engines, but they are unable to organize scientific knowledge alone. In this paper we introduce Galactica: a large language model that can store, combine and reason about scientific knowledge. We train on a large scientific corpus of papers, reference material, knowledge bases and many other sources. We outperform existing models on a range of scientific tasks. On technical knowledge probes such as LaTeX equations, Galactica outperforms the latest GPT-3 by 68.2% versus 49.0%. Galactica also performs well on reasoning, outperforming Chinchilla on mathematical MMLU by 41.3% to 35.7%, and PaLM 540B on MATH with a score of 20.4% versus 8.8%. It also sets a new state-of-the-art on downstream tasks such as PubMedQA and MedMCQA dev of 77.6% and 52.9%. And despite not being trained on a general corpus, Galactica outperforms BLOOM and OPT-175B on BIG-bench. We believe these results demonstrate the potential for language models as a new interface for science. We open source the model for the benefit of the scientific community.

Systems engineering, in particular in the automotive domain, needs to cope with the massively increasing numbers of requirements that arise during the development process. To guarantee a high product quality and make sure that functional safety standards such as ISO26262 are fulfilled, the exploitation of potentials of model-driven systems engineering in the form of automatic analyses, consistency checks, and tracing mechanisms is indispensable. However, the language in which requirements are written, and the tools needed to operate on them, are highly individual and require domain-specific tailoring. This hinders automated processing of requirements as well as the linking of requirements to models. Introducing formal requirement notations in existing projects leads to the challenge of translating masses of requirements and process changes on the one hand and to the necessity of the corresponding training for the requirements engineers. In this paper, based on the analysis of an open-source set of automotive requirements, we derive domain-specific language constructs helping us to avoid ambiguities in requirements and increase the level of formality. The main contribution is the adoption and evaluation of few-shot learning with large pretrained language models for the automated translation of informal requirements to structured languages such as a requirement DSL. We show that support sets of less than ten translation examples can suffice to few-shot train a language model to incorporate keywords and implement syntactic rules into informal natural language requirements.

Scaling language models improves performance but comes with significant computational costs. This paper proposes UL2R, a method that substantially improves existing language models and their scaling curves with a relatively tiny amount of extra compute. The key idea is to continue training a state-of-the-art large language model (e.g., PaLM) on a few more steps with UL2's mixture-of-denoiser objective. We show that, with almost negligible extra computational costs and no new sources of data, we are able to substantially improve the scaling properties of large language models on downstream metrics. In this paper, we continue training PaLM with UL2R, introducing a new set of models at 8B, 62B, and 540B scale which we call U-PaLM. Impressively, at 540B scale, we show an approximately 2x computational savings rate where U-PaLM achieves the same performance as the final PaLM 540B model at around half its computational budget (i.e., saving $\sim$4.4 million TPUv4 hours). We further show that this improved scaling curve leads to 'emergent abilities' on challenging BIG-Bench tasks -- for instance, U-PaLM does much better than PaLM on some tasks or demonstrates better quality at much smaller scale (62B as opposed to 540B). Overall, we show that U-PaLM outperforms PaLM on many few-shot setups, i.e., English NLP tasks (e.g., commonsense reasoning, question answering), reasoning tasks with chain-of-thought (e.g., GSM8K), multilingual tasks (MGSM, TydiQA), MMLU and challenging BIG-Bench tasks. Finally, we provide qualitative examples showing the new capabilities of U-PaLM for single and multi-span infilling.