Abstract:In this report, we present the latest model of the Gemini family, Gemini 1.5 Pro, a highly compute-efficient multimodal mixture-of-experts model 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. Gemini 1.5 Pro achieves near-perfect recall on long-context retrieval tasks across modalities, improves the state-of-the-art in long-document QA, long-video QA and long-context ASR, and matches or surpasses Gemini 1.0 Ultra's state-of-the-art performance across a broad set of benchmarks. Studying the limits of Gemini 1.5 Pro'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 2.1 (200k) and GPT-4 Turbo (128k). Finally, we highlight 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.
Abstract:We propose MusicRL, the first music generation system finetuned from human feedback. Appreciation of text-to-music models is particularly subjective since the concept of musicality as well as the specific intention behind a caption are user-dependent (e.g. a caption such as "upbeat work-out music" can map to a retro guitar solo or a techno pop beat). Not only this makes supervised training of such models challenging, but it also calls for integrating continuous human feedback in their post-deployment finetuning. MusicRL is a pretrained autoregressive MusicLM (Agostinelli et al., 2023) model of discrete audio tokens finetuned with reinforcement learning to maximise sequence-level rewards. We design reward functions related specifically to text-adherence and audio quality with the help from selected raters, and use those to finetune MusicLM into MusicRL-R. We deploy MusicLM to users and collect a substantial dataset comprising 300,000 pairwise preferences. Using Reinforcement Learning from Human Feedback (RLHF), we train MusicRL-U, the first text-to-music model that incorporates human feedback at scale. Human evaluations show that both MusicRL-R and MusicRL-U are preferred to the baseline. Ultimately, MusicRL-RU combines the two approaches and results in the best model according to human raters. Ablation studies shed light on the musical attributes influencing human preferences, indicating that text adherence and quality only account for a part of it. This underscores the prevalence of subjectivity in musical appreciation and calls for further involvement of human listeners in the finetuning of music generation models.
Abstract:Learning general-purpose representations from multisensor data produced by the omnipresent sensing systems (or IoT in general) has numerous applications in diverse use areas. Existing purely supervised end-to-end deep learning techniques depend on the availability of a massive amount of well-curated data, acquiring which is notoriously difficult but required to achieve a sufficient level of generalization on a task of interest. In this work, we leverage the self-supervised learning paradigm towards realizing the vision of continual learning from unlabeled inputs. We present a generalized framework named Sense and Learn for representation or feature learning from raw sensory data. It consists of eight auxiliary tasks that can learn high-level and broadly useful features entirely from unannotated data without any human involvement in the tedious labeling process. We demonstrate the efficacy of our approach on several publicly available datasets from different domains and in various settings, including linear separability, semi-supervised or few shot learning, and transfer learning. Our methodology achieves results that are competitive with the supervised approaches and close the gap through fine-tuning a network while learning the downstream tasks in most cases. In particular, we show that the self-supervised network can be utilized as initialization to significantly boost the performance in a low-data regime with as few as 5 labeled instances per class, which is of high practical importance to real-world problems. Likewise, the learned representations with self-supervision are found to be highly transferable between related datasets, even when few labeled instances are available from the target domains. The self-learning nature of our methodology opens up exciting possibilities for on-device continual learning.