In this report, we introduce the Gemini 1.5 family of models, representing the next generation of highly compute-efficient multimodal models capable of recalling and reasoning over fine-grained information from millions of tokens of context, including multiple long documents and hours of video and audio. The family includes two new models: (1) an updated Gemini 1.5 Pro, which exceeds the February version on the great majority of capabilities and benchmarks; (2) Gemini 1.5 Flash, a more lightweight variant designed for efficiency with minimal regression in quality. Gemini 1.5 models achieve near-perfect recall on long-context retrieval tasks across modalities, improve the state-of-the-art in long-document QA, long-video QA and long-context ASR, and match or surpass Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5's long-context ability, we find continued improvement in next-token prediction and near-perfect retrieval (>99%) up to at least 10M tokens, a generational leap over existing models such as Claude 3.0 (200k) and GPT-4 Turbo (128k). Finally, we highlight real-world use cases, such as Gemini 1.5 collaborating with professionals on completing their tasks achieving 26 to 75% time savings across 10 different job categories, as well as surprising new capabilities of large language models at the frontier; when given a grammar manual for Kalamang, a language with fewer than 200 speakers worldwide, the model learns to translate English to Kalamang at a similar level to a person who learned from the same content.

Writing complete stories is considered a hallmark display of human intelligence, and thus researchers in artificial intelligence (AI) and natural language generation (NLG) have long used it as a pinnacle task for their research (Klein et al., 1973; Meehan, 1977; Turner, 1993; Dehn, 1981; Liu and Singh, 2002; McIntyre and Lapata, 2009). Creative writing and storytelling present unique challenges for automatic language generation: story arcs extend over thousands of words, stories typically contain multiple characters with their own distinctive personas and voices, and well-written stories have an authorial voice that is consistent and identifiable. At the same time, lies and fabrications-common generation flaws which are a liability in tasks like machine translation and automatic summarization-can be an asset in the creative domain. In recent years, the field of NLG has progressed by leaps and bounds due to the development of neural language models capable of learning the structure of language by ingesting billions of written words (Chowdhery et al., 2022; Zhang et al., 2022; Brown et al., 2020). There has been considerable work in applying these advancements toward the development of AI-powered tools for creative writing, but nearly all previous research in this space has evaluated their methods either with amateur writers or with crowd workers paid to assess performance on narrowly defined tasks (Clark et al., 2018; Roemmele and Gordon, 2015; Nichols et al., 2020). While these sorts of evaluations are valuable as preliminary assessments, we believe it is also crucial to solicit feedback from actual domain experts in creative writing: professional writers, educators, and language experts. Skilled writers comprise a unique user group with a different set of needs and expectations than amateurs.

NLP researchers need more, higher-quality text datasets. Human-labeled datasets are expensive to collect, while datasets collected via automatic retrieval from the web such as WikiBio are noisy and can include undesired biases. Moreover, data sourced from the web is often included in datasets used to pretrain models, leading to inadvertent cross-contamination of training and test sets. In this work we introduce a novel method for efficient dataset curation: we use a large language model to provide seed generations to human raters, thereby changing dataset authoring from a writing task to an editing task. We use our method to curate SynthBio - a new evaluation set for WikiBio - composed of structured attribute lists describing fictional individuals, mapped to natural language biographies. We show that our dataset of fictional biographies is less noisy than WikiBio, and also more balanced with respect to gender and nationality.

Transformer architectures show significant promise for natural language processing. Given that a single pretrained model can be fine-tuned to perform well on many different tasks, these networks appear to extract generally useful linguistic features. A natural question is how such networks represent this information internally. This paper describes qualitative and quantitative investigations of one particularly effective model, BERT. At a high level, linguistic features seem to be represented in separate semantic and syntactic subspaces.

Transformer architectures show significant promise for natural language processing. Given that a single pretrained model can be fine-tuned to perform well on many different tasks, these networks appear to extract generally useful linguistic features. A natural question is how such networks represent this information internally. This paper describes qualitative and quantitative investigations of one particularly effective model, BERT. At a high level, linguistic features seem to be represented in separate semantic and syntactic subspaces. We find evidence of a fine-grained geometric representation of word senses. We also present empirical descriptions of syntactic representations in both attention matrices and individual word embeddings, as well as a mathematical argument to explain the geometry of these representations.