Recent work claims that large language models display emergent abilities, abilities not present in smaller-scale models that are present in larger-scale models. What makes emergent abilities intriguing is two-fold: their sharpness, transitioning seemingly instantaneously from not present to present, and their unpredictability, appearing at seemingly unforeseeable model scales. Here, we present an alternative explanation for emergent abilities: that for a particular task and model family, when analyzing fixed model outputs, one can choose a metric which leads to the inference of an emergent ability or another metric which does not. Thus, our alternative suggests that existing claims of emergent abilities are creations of the researcher's analyses, not fundamental changes in model behavior on specific tasks with scale. We present our explanation in a simple mathematical model, then test it in three complementary ways: we (1) make, test and confirm three predictions on the effect of metric choice using the InstructGPT/GPT-3 family on tasks with claimed emergent abilities, (2) make, test and confirm two predictions about metric choices in a meta-analysis of emergent abilities on BIG-Bench; and (3) show how similar metric decisions suggest apparent emergent abilities on vision tasks in diverse deep network architectures (convolutional, autoencoder, transformers). In all three analyses, we find strong supporting evidence that emergent abilities may not be a fundamental property of scaling AI models.
相關內容
ACM/IEEE第23屆模型驅動工程語言和系統國際會議，是模型驅動軟件和系統工程的首要會議系列，由ACM-SIGSOFT和IEEE-TCSE支持組織。自1998年以來，模型涵蓋了建模的各個方面，從語言和方法到工具和應用程序。模特的參加者來自不同的背景，包括研究人員、學者、工程師和工業專業人士。MODELS 2019是一個論壇，參與者可以圍繞建模和模型驅動的軟件和系統交流前沿研究成果和創新實踐經驗。今年的版本將為建模社區提供進一步推進建模基礎的機會，并在網絡物理系統、嵌入式系統、社會技術系統、云計算、大數據、機器學習、安全、開源等新興領域提出建模的創新應用以及可持續性。
官網鏈接： ·
MoDELS
·
Performer
·
ImageNet (數據集)
·
Learning
·
We argue that there are many notions of 'similarity' and that models, like humans, should be able to adapt to these dynamically. This contrasts with most representation learning methods, supervised or self-supervised, which learn a fixed embedding function and hence implicitly assume a single notion of similarity. For instance, models trained on ImageNet are biased towards object categories, while a user might prefer the model to focus on colors, textures or specific elements in the scene. In this paper, we propose the GeneCIS ('genesis') benchmark, which measures models' ability to adapt to a range of similarity conditions. Extending prior work, our benchmark is designed for zero-shot evaluation only, and hence considers an open-set of similarity conditions. We find that baselines from powerful CLIP models struggle on GeneCIS and that performance on the benchmark is only weakly correlated with ImageNet accuracy, suggesting that simply scaling existing methods is not fruitful. We further propose a simple, scalable solution based on automatically mining information from existing image-caption datasets. We find our method offers a substantial boost over the baselines on GeneCIS, and further improves zero-shot performance on related image retrieval benchmarks. In fact, though evaluated zero-shot, our model surpasses state-of-the-art supervised models on MIT-States. Project page at //sgvaze.github.io/genecis/.
With their increasing size, Large language models (LLMs) are becoming increasingly good at language understanding tasks. But even with high performance on specific downstream task, LLMs fail at simple linguistic tests for negation or quantifier understanding. Previous work on testing capability of LLMs on understanding quantifiers suggest that as the size of the models increase, they get better at understanding most-type quantifiers but get increasingly worse at understanding few-type quantifiers, thus presenting a case of an inverse-scaling law. In this paper, we question the claims of inverse scaling of few-type quantifier understanding in LLMs and show that it is a result of inappropriate testing methodology. We also present alternate methods to measure quantifier comprehension in LLMs and show that as the size of the models increase, these behaviours are different from what is shown in previous research. LLMs are consistently able to understand the difference between the meaning of few-type and most-type quantifiers, but when a quantifier is added to phrase, LLMs do not always take into account the meaning of the quantifier. We in fact see an inverse scaling law for most-type quantifiers, which is contrary to human psycho-linguistic experiments and previous work, where the model's understanding of most-type quantifier gets worse as the model size increases. We do this evaluation on models ranging from 125M-175B parameters, which suggests that LLMs do not do as well as expected with quantifiers and statistical co-occurrence of words still takes precedence over word meaning.
Recursion is one of the hallmarks of human language. While many design features of language have been shown to exist in animal communication systems, recursion has not. Previous research shows that GPT-4 is the first large language model (LLM) to exhibit metalinguistic abilities (Begu\v{s}, D\k{a}bkowski, and Rhodes 2023). Here, we propose several prompt designs aimed at eliciting and analyzing recursive behavior in LLMs, both linguistic and non-linguistic. We demonstrate that when explicitly prompted, GPT-4 can both produce and analyze recursive structures. Thus, we present one of the first studies investigating whether meta-linguistic awareness of recursion -- a uniquely human cognitive property -- can emerge in transformers with a high number of parameters such as GPT-4.
Better understanding of Large Language Models' (LLMs) legal analysis abilities can contribute to improving the efficiency of legal services, governing artificial intelligence, and leveraging LLMs to identify inconsistencies in law. This paper explores LLM capabilities in applying tax law. We choose this area of law because it has a structure that allows us to set up automated validation pipelines across thousands of examples, requires logical reasoning and maths skills, and enables us to test LLM capabilities in a manner relevant to real-world economic lives of citizens and companies. Our experiments demonstrate emerging legal understanding capabilities, with improved performance in each subsequent OpenAI model release. We experiment with retrieving and utilising the relevant legal authority to assess the impact of providing additional legal context to LLMs. Few-shot prompting, presenting examples of question-answer pairs, is also found to significantly enhance the performance of the most advanced model, GPT-4. The findings indicate that LLMs, particularly when combined with prompting enhancements and the correct legal texts, can perform at high levels of accuracy but not yet at expert tax lawyer levels. As LLMs continue to advance, their ability to reason about law autonomously could have significant implications for the legal profession and AI governance.
The high computational and memory requirements of large language model (LLM) inference make it feasible only with multiple high-end accelerators. Motivated by the emerging demand for latency-insensitive tasks with batched processing, this paper initiates the study of high-throughput LLM inference using limited resources, such as a single commodity GPU. We present FlexGen, a high-throughput generation engine for running LLMs with limited GPU memory. FlexGen can be flexibly configured under various hardware resource constraints by aggregating memory and computation from the GPU, CPU, and disk. By solving a linear programming problem, it searches for efficient patterns to store and access tensors. FlexGen further compresses the weights and the attention cache to 4 bits with negligible accuracy loss. These techniques enable FlexGen to have a larger space of batch size choices and thus significantly increase maximum throughput. As a result, when running OPT-175B on a single 16GB GPU, FlexGen achieves significantly higher throughput compared to state-of-the-art offloading systems, reaching a generation throughput of 1 token/s for the first time with an effective batch size of 144. On the HELM benchmark, FlexGen can benchmark a 30B model with a 16GB GPU on 7 representative sub-scenarios in 21 hours. The code is available at //github.com/FMInference/FlexGen
Language has a strong influence on our perceptions of time and rewards. This raises the question of whether large language models, when asked in different languages, show different preferences for rewards over time and if their choices are similar to those of humans. In this study, we analyze the responses of GPT-3.5 (hereafter referred to as GPT) to prompts in multiple languages, exploring preferences between smaller, sooner rewards and larger, later rewards. Our results show that GPT displays greater patience when prompted in languages with weak future tense references (FTR), such as German and Mandarin, compared to languages with strong FTR, like English and French. These findings are consistent with existing literature and suggest a correlation between GPT's choices and the preferences of speakers of these languages. However, further analysis reveals that the preference for earlier or later rewards does not systematically change with reward gaps, indicating a lexicographic preference for earlier payments. While GPT may capture intriguing variations across languages, our findings indicate that the choices made by these models do not correspond to those of human decision-makers.
Molecule discovery plays a crucial role in various scientific fields, advancing the design of tailored materials and drugs. Traditional methods for molecule discovery follow a trial-and-error process, which are both time-consuming and costly, while computational approaches such as artificial intelligence (AI) have emerged as revolutionary tools to expedite various tasks, like molecule-caption translation. Despite the importance of molecule-caption translation for molecule discovery, most of the existing methods heavily rely on domain experts, require excessive computational cost, and suffer from poor performance. On the other hand, Large Language Models (LLMs), like ChatGPT, have shown remarkable performance in various cross-modal tasks due to their great powerful capabilities in natural language understanding, generalization, and reasoning, which provides unprecedented opportunities to advance molecule discovery. To address the above limitations, in this work, we propose a novel LLMs-based framework (\textbf{MolReGPT}) for molecule-caption translation, where a retrieval-based prompt paradigm is introduced to empower molecule discovery with LLMs like ChatGPT without fine-tuning. More specifically, MolReGPT leverages the principle of molecular similarity to retrieve similar molecules and their text descriptions from a local database to ground the generation of LLMs through in-context few-shot molecule learning. We evaluate the effectiveness of MolReGPT via molecule-caption translation, which includes molecule understanding and text-based molecule generation. Experimental results show that MolReGPT outperforms fine-tuned models like MolT5-base without any additional training. To the best of our knowledge, MolReGPT is the first work to leverage LLMs in molecule-caption translation for advancing molecule discovery.
The impressive recent performance of large language models has led many to wonder to what extent they can serve as models of general intelligence or are similar to human cognition. We address this issue by applying GPT-3 and GPT-4 to a classic problem in human inductive reasoning known as property induction. Over two experiments, we elicit human judgments on a range of property induction tasks spanning multiple domains. Although GPT-3 struggles to capture many aspects of human behaviour, GPT-4 is much more successful: for the most part, its performance qualitatively matches that of humans, and the only notable exception is its failure to capture the phenomenon of premise non-monotonicity. Overall, this work not only demonstrates that property induction is an interesting skill on which to compare human and machine intelligence, but also provides two large datasets that can serve as suitable benchmarks for future work in this vein.
As LLMs become commonplace, machine-generated text has the potential to flood the internet with spam, social media bots, and valueless content. Watermarking is a simple and effective strategy for mitigating such harms by enabling the detection and documentation of LLM-generated text. Yet a crucial question remains: How reliable is watermarking in realistic settings in the wild? There, watermarked text may be modified to suit a user's needs, or entirely rewritten to avoid detection. We study the robustness of watermarked text after it is re-written by humans, paraphrased by a non-watermarked LLM, or mixed into a longer hand-written document. We find that watermarks remain detectable even after human and machine paraphrasing. While these attacks dilute the strength of the watermark, paraphrases are statistically likely to leak n-grams or even longer fragments of the original text, resulting in high-confidence detections when enough tokens are observed. For example, after strong human paraphrasing the watermark is detectable after observing 800 tokens on average, when setting a 1e-5 false positive rate. We also consider a range of new detection schemes that are sensitive to short spans of watermarked text embedded inside a large document, and we compare the robustness of watermarking to other kinds of detectors.
Reasoning is a fundamental aspect of human intelligence that plays a crucial role in activities such as problem solving, decision making, and critical thinking. In recent years, large language models (LLMs) have made significant progress in natural language processing, and there is observation that these models may exhibit reasoning abilities when they are sufficiently large. However, it is not yet clear to what extent LLMs are capable of reasoning. This paper provides a comprehensive overview of the current state of knowledge on reasoning in LLMs, including techniques for improving and eliciting reasoning in these models, methods and benchmarks for evaluating reasoning abilities, findings and implications of previous research in this field, and suggestions on future directions. Our aim is to provide a detailed and up-to-date review of this topic and stimulate meaningful discussion and future work.
注冊地址： 北京市海淀區羊坊店路18號2幢3層301-191