AI Technology & Industry Review
Technical review of the newest machine intelligence research.
In a new paper How Does Generative Retrieval Scale to Millions of Passages? a research team from Google Research and University of Waterloo performs the first empirical study of generative retrieval across various corpus scales, even scaling up to the entire MS MARCO passage ranking task that contains 8.8M passages, aiming to provide insights on scaling generative retrieval to millions of passages.
In the new paper DoReMi: Optimizing Data Mixtures Speeds Up Language Model Pretraining, a research team from Google and Stanford University introduces Domain Reweighting with Minimax Optimization (DoReMi), a domain weight optimization strategy that leverages distributionally robust optimization (DRO) to substantially speed up effective language model pretraining.
In the new paper Large Language Models as Tool Makers, a research team from Google DeepMind, Princeton University and Stanford University presents LATM (large language models as tool makers), a closed-loop framework that enables LLMs to create their own reusable tools to boost efficiency and enhance their problem-solving capabilities.
In the new paper READ: Recurrent Adaptation of Large Transformers, a Meta AI research team proposes REcurrent ADaption (READ), a lightweight and memory-efficient fine-tuning approach that achieves a 56 percent reduction in memory consumption and an 84 percent reduction in GPU use.
In the new paper Scaling Speech Technology to 1,000+ Languages, a Meta AI research team launches the company’s Massively Multilingual Speech (MMS) project, which aims to dramatically expand access to speech technologies.
In the new paper ONE-PEACE: Exploring One General Representation Model Toward Unlimited Modalities, a research team from Alibaba Group’s DAMO Academy and the Huazhong University of Science and Technology releases ONE-PEACE, a highly extensible model that can align and integrate representations across vision, audio, and language modalities; opening a path toward the creation of a general representation model for unlimited modalities.
In the paper PaLM 2 Technical Report, Google details the modelling advances, data improvements and scaling insights that have enabled its new PaLM2 large language model to achieve state-of-the-art performance.
In the new paper MegaByte: Predicting Million-Byte Sequences with Multiscale Transformers, a Meta AI research team presents MegaByte, a multiscale decoder architecture that enables million-byte sequence modelling.
In the new paper CodeT5+: Open Code Large Language Models for Code Understanding and Generation, a Salesforce AI Research team presents CodeT5+, a novel family of encoder-decoder code foundation large language models that can be flexibly adapted to a wide range of code understanding and generation tasks and outperform various code-related benchmarks.
In the new paper StarCoder: May the Source Be With You!, the BigCode community releases StarCoder and StarCoderBase, 15.5B parameter open-access large language models (LLMs) trained on 80+ programming languages. StarCoderBase outperforms all multi-programming-language code LLMs, and StarCoder surpasses all models fine-tuned on Python.
In the new paper VideoChat: Chat-Centric Video Understanding, a research team from Shanghai AI Laboratory, Nanjing University, the University of Hong Kong, and the Chinese Academy of Sciences presents VideoChat, a groundbreaking end-to-end chat-centric video understanding system that leverages state-of-the-art video and language models to improve spatiotemporal reasoning, event localization, and causal relationship inference.
In the new paper Shap·E: Generating Conditional 3D Implicit Functions, an OpenAI research team proposes Shap·E, a conditional generative model that leverages implicit neural representations to produce complex and diverse 3D assets.
In the new paper ZipIt! Merging Models from Different Tasks Without Training, a Georgia Tech research team proposes ZipIt!, a general method that exploits redundant features to combine two or more models with the same architecture but trained on different tasks into one multi-task model without additional training.
In the new paper Automatic Prompt Optimization with “Gradient Descent” and Beam Search, a Microsoft research team presents Automatic Prompt Optimization, a simple and general prompt optimization algorithm that automatically improves prompts for large language models, significantly reducing the time and energy spent on manual prompting approaches.
In the new paper Finding Neurons in a Haystack: Case Studies with Sparse Probing, a research team from MIT, Harvard University and Northeastern University proposes sparse probing, a technique that probes over 100 features to precisely localize the neurons in large language models that are relevant to a specific feature or concept.
In the new paper Unlimiformer: Long-Range Transformers With Unlimited Length Input, a Carnegie Mellon University research team presents a general approach for improving model performance by augmenting pretrained encoder-decoder transformers with an external datastore to permit inputs of unbounded length.
In the new paper ResiDual: Transformer With Dual Residual Connections, a team from Microsoft Research, Microsoft Azure Translation, and Renmin University of China proposes ResiDual, a novel transformer architecture that fuses the connections in post-layer normalization and pre-layer normalization to exploit the benefits of both while also addressing their limitations.
In the new paper Dissecting Recall of Factual Associations in Auto-Regressive Language Models, a team from Google DeepMind, Tel Aviv University and Google Research investigates how factual associations are stored and extracted internally in transformer-based language models and provides insights on how such models’ factual predictions are formed.
In the new paper Towards Realistic Generative 3D Face Models, a research team from Carnegie Mellon University and Meta proposes a 3D controllable generative model capable of generating high-resolution textures and capturing high-frequency details in facial geometry. Their proposed AlbedoGAN outperforms state-of-the-art baselines in facial shape reconstruction.
In the new paper WizardLM: Empowering Large Language Models to Follow Complex Instructions, a research team from Microsoft and Peking University presents Evol-Instruct, a novel approach that leverages LLMs to automatically generate large amounts of instruction data with varying levels of complexity. In human evaluations, the team’s resulting WizardLM model’s generated instructions were judged superior to human-created instruction datasets.
In the new paper FastRLAP: A System for Learning High-Speed Driving via Deep RL and Autonomous Practicing, a UC Berkeley research team proposes FastRLAP (Fast Reinforcement Learning via Autonomous Practicing), a system that autonomously practices in the real world and learns aggressive maneuvers to enable effective high-speed driving.
In the new paper NaturalSpeech 2: Latent Diffusion Models are Natural and Zero-Shot Speech and Singing Synthesizers, a Microsoft team introduces NaturalSpeech 2, a TTS system with latent diffusion models for natural and strong zero-shot voice synthesis that captures expressive prosodies with superior robustness.
In the new paper DETRs Beat YOLOs on Real-Time Object Detection, a Baidu Inc. research team presents Real-Time Detection Transformer (RT-DETR), a real-time end-to-end object detector that leverages a hybrid encoder and novel IoU-aware query selection to address inference speed delay issues. RT-DETR outperforms YOLO object detectors in both accuracy and speed.
In the new paper DiffFit: Unlocking Transferability of Large Diffusion Models via Simple Parameter-Efficient Fine-Tuning, a Huawei Noah’s Ark Lab research team introduces DiffFit, a parameter-efficient fine-tuning technique that enables fast adaptation to new domains for diffusion image generation. Compared to full fine-tuning approaches, DiffFit achieves 2x training speed-ups while using only ~0.12 percent of trainable parameters.
In the new paper Meta-Learned Models of Cognition, a team from the Max Planck Institute for Biological Cybernetics (Max-Planck-Gesellschaft, MPG) and DeepMind proposes the establishment of a research program focused on meta-learned models of cognition. The team cites machine learning papers demonstrating how meta-learning can be used to construct Bayes-optimal learning algorithms and suggests it can significantly expand the scope of the rational analysis of cognition.
In the new paper SpectFormer: Frequency and Attention Is What You Need in a Vision Transformer, a research team from Microsoft and the University of Bath proposes Spectformer, a novel transformer architecture that combines spectral and multi-headed attention layers to better capture appropriate feature representations and improve performance.
In the new paper Teaching Large Language Models to Self-Debug, a Google Research and UC Berkeley team presents Self-Debugging, a framework that teaches large language models to debug their own predicted code via few-shot demonstrations and improves baseline accuracy by up to 12 percent.
In the new paper Inference with Reference: Lossless Acceleration of Large Language Models, a Microsoft research team proposes LLMA, an inference-with-reference decoding mechanism that achieves up to 2x lossless speed-ups with identical generation results by exploiting the overlaps between LLM outputs and references.
In the new paper Generative Agents: Interactive Simulacra of Human Behavior, a team from Stanford University and Google Research presents agents that draw on generative models to simulate both individual and emergent group behaviours that are humanlike and based on their changing experiences and environment.
In the new paper Pix2Video: Video Editing Using Image Diffusion, an Adobe Research and University College London team presents Pix2Video, a framework for realistic text-guided video editing using a pretrained image diffusion model.
In the new paper Instruct-NeRF2NeRF: Editing 3D Scenes With Instructions, a UC Berkeley research team presents Instruct-NeRF2NeRF, an approach for editing 3D NeRF scenes through natural language text instructions. The proposed method can edit large-scale, real-world 3D scenes with improved ease of use and realism.
In the new paper TPU v4: An Optically Reconfigurable Supercomputer for Machine Learning with Hardware Support, a Google Research team presents TPU v4, the company’s latest supercomputer. TPU v4 is ten times faster than v3 and 1.2–1.7x faster than Nvidia A100 GPUs while using 1.3x–1.9x less power.
In the new paper Towards Healthy AI: Large Language Models Need Therapists Too, a team from Columbia University and IBM Research proposes SafeguardGPT, a framework that incorporates psychotherapy and reinforcement learning to correct the potentially harmful behaviours of AI chatbots.
In the new paper BloombergGPT: A Large Language Model for Finance, a research team from Bloomberg and Johns Hopkins University presents BloombergGPT, a 50 billion parameter language model trained on a 700 billion token dataset that significantly outperforms current benchmark models on financial tasks.
In the new paper TaskMatrix.AI: Completing Tasks by Connecting Foundation Models with Millions of APIs, a Microsoft research team proposes TaskMatrix.AI, a novel ecosystem that connects foundation models with millions of existing models and system APIs to build a “super-AI” capable of addressing a wide range of digital and physical tasks.
A recent tech demo called “Living Chang’an City” has been garnering attention. In this video, AI-powered NPCs can be seen roaming the streets of Chang’an City, each possessing unique identities and short-term and long-term goals. They engage in various life-like interactions, such as chatting, shopping and even falling in love.
Colossal-AI open sources a complete RLHF pipeline that includes supervised data collection, supervised fine-tuning, reward model training, and reinforcement learning fine-tuning, based on the LLaMA pre-trained model, and shares ColossalChat, the most practical open-source project that closely resembles the original ChatGPT technical solution!
A Google Research team addresses transformers’ input sequence limitations in the new paper CoLT5: Faster Long-Range Transformers with Conditional Computation, proposing CoLT5 (Conditional LongT5), a family of models that applies a novel conditional computation approach for higher quality and faster long-input processing of up to 64,000 tokens.
In the new paper Sparks of Artificial General Intelligence: Early Experiments with GPT-4, a Microsoft Research team investigates GPT-4, demonstrating its ability to achieve human-level performance on novel and difficult tasks in domains ranging from mathematics and coding to vision, medicine, law and psychology; and proposing it as an early version of an AGI system.
In the new paper GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models, a research team from OpenAI, OpenResearch, and the University of Pennsylvania investigates the potential impact of LLMs like GPT on the US labour market, shedding light on the economic, social, and policy implications.