Negligible in Size, Significant in Effect: On Scale Vectors in Large Language Models

Normalization layers in modern large language models (LLMs) consist of a deterministic normalization operation and a learnable scale vector. While the normalization operation has been extensively studied, the scale vector remains poorly understood despite its ubiquitous use. In this work, we present a systematic study of scale vectors in LLMs from the perspectives of expressivity, optimization, and architectural structure. First, we show empirically that although scale vectors constitute only a negligible fraction of model parameters, removing them substantially degrades LLM pre-training. Our theory further shows that, in Pre-Norm architectures, scale vectors do not increase expressivity; instead, they improve optimization through a self-amplifying preconditioning effect on subsequent linear mappings. Second, we investigate the role of weight decay for scale vectors. By distinguishing Input-Norm and Output-Norm layers, we theoretically show that weight decay is beneficial for the former but harmful for the latter, due to their distinct roles in optimization and expressivity. Third, motivated by this understanding, we propose three lightweight and complementary improvements to scale vectors: branch-specific heterogeneity, improved placement around linear mappings, and magnitude-direction reparameterization. Both theory and experiments show that each improvement yields consistent gains. Finally, we combine these improvements into a unified scale-vector strategy and evaluate it through extensive LLM pre-training experiments on dense and mixture-of-experts models ranging from 0.12B to 2B parameters, across multiple optimizers and learning rate schedules, under industrial-scale token budgets. The unified strategy consistently achieves lower terminal loss than well-tuned baselines and exhibits more favorable scaling behavior, while adding negligible parameter and computational overhead.

More Expressive Feedforward Layers: Part I. Token-Adaptive Mixing of Activations

Feedforward network (FFN) layers account for a large fraction of parameters and nonlinear expressivity in Transformer-based large language models (LLMs). Despite the evolution from ReLU and GELU to gated variants such as SwiGLU, most FFN designs still use a single fixed activation function, applying the same nonlinear transformation to all tokens. In this work, we propose Mixture of Activations (MoA), a token-adaptive FFN design that mixes a dictionary of activation functions using lightweight input-dependent gates while sharing the same linear projections. As an input-independent counterpart, we also introduce learnable activations (LA), which form linear combinations of activation functions for both ReLU-type and SwiGLU-type FFNs. Theoretically, we establish strict finite-width expressive separations among fixed-activation FFNs, LA, and MoA: LA strictly contains fixed-activation FFNs, while MoA strictly contains LA, with the additional expressivity arising from input-dependent nonlinear hybridization. Empirically, we evaluate MoA through extensive pre-training experiments on dense and MoE language models ranging from 0.12B to 2B parameters under different token budgets, optimizers, and learning rate schedules. MoA consistently achieves lower terminal loss and exhibits more favorable scaling behavior than well-tuned baselines, with minimal parameter and computational overhead. These results suggest that token-adaptive activation mixing is a simple and effective mechanism for improving FFN expressivity in LLMs.

What happens when reviewers receive AI feedback in their reviews?

AI is reshaping academic research, yet its role in peer review remains polarising and contentious. Advocates see its potential to reduce reviewer burden and improve quality, while critics warn of risks to fairness, accountability, and trust. At ICLR 2025, an official AI feedback tool was deployed to provide reviewers with post-review suggestions. We studied this deployment through surveys and interviews, investigating how reviewers engaged with the tool and perceived its usability and impact. Our findings surface both opportunities and tensions when AI augments in peer review. This work contributes the first empirical evidence of such an AI tool in a live review process, documenting how reviewers respond to AI-generated feedback in a high-stakes review context. We further offer design implications for AI-assisted reviewing that aim to enhance quality while safeguarding human expertise, agency, and responsibility.

Artificial Hippocampus Networks for Efficient Long-Context Modeling

Long-sequence modeling faces a fundamental trade-off between the efficiency of compressive fixed-size memory in RNN-like models and the fidelity of lossless growing memory in attention-based Transformers. Inspired by the Multi-Store Model in cognitive science, we introduce a memory framework of artificial neural networks. Our method maintains a sliding window of the Transformer's KV cache as lossless short-term memory, while a learnable module termed Artificial Hippocampus Network (AHN) recurrently compresses out-of-window information into a fixed-size compact long-term memory. To validate this framework, we instantiate AHNs using modern RNN-like architectures, including Mamba2, DeltaNet, and Gated DeltaNet. Extensive experiments on long-context benchmarks LV-Eval and InfiniteBench demonstrate that AHN-augmented models consistently outperform sliding window baselines and achieve performance comparable or even superior to full-attention models, while substantially reducing computational and memory requirements. For instance, augmenting the Qwen2.5-3B-Instruct with AHNs reduces inference FLOPs by 40.5% and memory cache by 74.0%, while improving its average score on LV-Eval (128k sequence length) from 4.41 to 5.88. Code is available at: https://github.com/ByteDance-Seed/AHN.

Understanding Transformer from the Perspective of Associative Memory

In this paper, we share our reflections and insights on understanding Transformer architectures through the lens of associative memory--a classic psychological concept inspired by human cognition. We start with the basics of associative memory (think simple linear attention) and then dive into two dimensions: Memory Capacity: How much can a Transformer really remember, and how well? We introduce retrieval SNR to measure this and use a kernel perspective to mathematically reveal why Softmax Attention is so effective. We also show how FFNs can be seen as a type of associative memory, leading to insights on their design and potential improvements. Memory Update: How do these memories learn and evolve? We present a unified framework for understanding how different Transformer variants (like DeltaNet and Softmax Attention) update their "knowledge base". This leads us to tackle two provocative questions: 1. Are Transformers fundamentally limited in what they can express, and can we break these barriers? 2. If a Transformer had infinite context, would it become infinitely intelligent? We want to demystify Transformer architecture, offering a clearer understanding of existing designs. This exploration aims to provide fresh insights and spark new avenues for Transformer innovation.

Emerging Properties in Unified Multimodal Pretraining

Unifying multimodal understanding and generation has shown impressive capabilities in cutting-edge proprietary systems. In this work, we introduce BAGEL, an open0source foundational model that natively supports multimodal understanding and generation. BAGEL is a unified, decoder0only model pretrained on trillions of tokens curated from large0scale interleaved text, image, video, and web data. When scaled with such diverse multimodal interleaved data, BAGEL exhibits emerging capabilities in complex multimodal reasoning. As a result, it significantly outperforms open-source unified models in both multimodal generation and understanding across standard benchmarks, while exhibiting advanced multimodal reasoning abilities such as free-form image manipulation, future frame prediction, 3D manipulation, and world navigation. In the hope of facilitating further opportunities for multimodal research, we share the key findings, pretraining details, data creation protocal, and release our code and checkpoints to the community. The project page is at https://bagel-ai.org/

Seed1.5-VL Technical Report

We present Seed1.5-VL, a vision-language foundation model designed to advance general-purpose multimodal understanding and reasoning. Seed1.5-VL is composed with a 532M-parameter vision encoder and a Mixture-of-Experts (MoE) LLM of 20B active parameters. Despite its relatively compact architecture, it delivers strong performance across a wide spectrum of public VLM benchmarks and internal evaluation suites, achieving the state-of-the-art performance on 38 out of 60 public benchmarks. Moreover, in agent-centric tasks such as GUI control and gameplay, Seed1.5-VL outperforms leading multimodal systems, including OpenAI CUA and Claude 3.7. Beyond visual and video understanding, it also demonstrates strong reasoning abilities, making it particularly effective for multimodal reasoning challenges such as visual puzzles. We believe these capabilities will empower broader applications across diverse tasks. In this report, we mainly provide a comprehensive review of our experiences in building Seed1.5-VL across model design, data construction, and training at various stages, hoping that this report can inspire further research. Seed1.5-VL is now accessible at https://www.volcengine.com/ (Volcano Engine Model ID: doubao-1-5-thinking-vision-pro-250428)

Sniff AI: Is My 'Spicy' Your 'Spicy'? Exploring LLM's Perceptual Alignment with Human Smell Experiences

Aligning AI with human intent is important, yet perceptual alignment-how AI interprets what we see, hear, or smell-remains underexplored. This work focuses on olfaction, human smell experiences. We conducted a user study with 40 participants to investigate how well AI can interpret human descriptions of scents. Participants performed "sniff and describe" interactive tasks, with our designed AI system attempting to guess what scent the participants were experiencing based on their descriptions. These tasks evaluated the Large Language Model's (LLMs) contextual understanding and representation of scent relationships within its internal states - high-dimensional embedding space. Both quantitative and qualitative methods were used to evaluate the AI system's performance. Results indicated limited perceptual alignment, with biases towards certain scents, like lemon and peppermint, and continued failing to identify others, like rosemary. We discuss these findings in light of human-AI alignment advancements, highlighting the limitations and opportunities for enhancing HCI systems with multisensory experience integration.

Exploring Human-AI Perception Alignment in Sensory Experiences: Do LLMs Understand Textile Hand?

Aligning large language models (LLMs) behaviour with human intent is critical for future AI. An important yet often overlooked aspect of this alignment is the perceptual alignment. Perceptual modalities like touch are more multifaceted and nuanced compared to other sensory modalities such as vision. This work investigates how well LLMs align with human touch experiences using the "textile hand" task. We created a "Guess What Textile" interaction in which participants were given two textile samples -- a target and a reference -- to handle. Without seeing them, participants described the differences between them to the LLM. Using these descriptions, the LLM attempted to identify the target textile by assessing similarity within its high-dimensional embedding space. Our results suggest that a degree of perceptual alignment exists, however varies significantly among different textile samples. For example, LLM predictions are well aligned for silk satin, but not for cotton denim. Moreover, participants didn't perceive their textile experiences closely matched by the LLM predictions. This is only the first exploration into perceptual alignment around touch, exemplified through textile hand. We discuss possible sources of this alignment variance, and how better human-AI perceptual alignment can benefit future everyday tasks.

MiliPoint: A Point Cloud Dataset for mmWave Radar

Millimetre-wave (mmWave) radar has emerged as an attractive and cost-effective alternative for human activity sensing compared to traditional camera-based systems. mmWave radars are also non-intrusive, providing better protection for user privacy. However, as a Radio Frequency (RF) based technology, mmWave radars rely on capturing reflected signals from objects, making them more prone to noise compared to cameras. This raises an intriguing question for the deep learning community: Can we develop more effective point set-based deep learning methods for such attractive sensors? To answer this question, our work, termed MiliPoint, delves into this idea by providing a large-scale, open dataset for the community to explore how mmWave radars can be utilised for human activity recognition. Moreover, MiliPoint stands out as it is larger in size than existing datasets, has more diverse human actions represented, and encompasses all three key tasks in human activity recognition. We have also established a range of point-based deep neural networks such as DGCNN, PointNet++ and PointTransformer, on MiliPoint, which can serve to set the ground baseline for further development.