AI·LLM-as-judge scores fluctuate across evaluators even with fixed candidate responses, treating the issue as measurement validity. Across four datasets, scaling Qwen3 from 1.7B to 32B parameters yields only limited adjacent gains, while MiniMax M2-to-M2.7 API upgrades show none; stronger judges reduce but do not eliminate position and verbosity biases, with repeated juries offering little benefit under correlated errors.
Search
What are you looking for?
11 results for “scaling”
AI·Hidden Decoding expands each token into n independent streams during continued pretraining of a fixed Transformer backbone, using separate embedding tables and retaining intermediate KV caches for latent computation; Stream-Factorized Attention limits cross-stream mixing to reduce costs. Experiments show frontier 80B and 617B MoE models improve on all benchmarks over matched baselines.
AI·The paper proposes UltraX, a method for refining large-scale pre-training data using adaptive programmatic editing to improve LLM quality when scaling data yields diminishing returns.
AI·Follow-up post details data-engineering optimizations that scaled a SQLite/FTS5 patent database from 3.5M to 5.36M records while classifying them with Nemotron 9B on an RTX 5090.
Despite the recent promise in robot control, video generative models suffer from a domain mismatch due to their primary focus on content creation. For example, their design inherently prioritizes visual fidelity and creativity over computational efficiency and physical realism. In this work, we present LingBot-Video, a DiT-based video pretraining paradigm specifically tailored for embodied intelligence. From the architecture perspective, we adopt the Mixture-of-Experts (MoE), instead of dense, …
Modern LLMs are increasingly deployed in long-context applications such as retrieval-augmented generation, repository-level coding, and agentic workflows whose accumulated reasoning and tool traces routinely push the input an order of magnitude past the pretraining window, making zero-shot context extension the dominant deployment path for open-weight checkpoints. Most existing zero-shot methods fix a single rescaling factor up front, so an aggressive factor sacrifices short-context fidelity wh…
Linear attention models allow a fixed state size and a fixed amount of compute per token. However, due to their limited state size, linear attention models fall behind in long-context recall compared to softmax-attention-based transformer architectures. Increasing the state size of linear attention improves recall performance but at the cost of higher FLOPs. In this work, we introduce Sparse Delta Memory (SDM), an architecture that scales the hidden state of gated linear RNNs to orders of magni…
Humans can navigate an unfamiliar city and gradually form a coherent spatial mental map spanning tens of square kilometers. Can AI build spatial representations at a comparable scale? Although recent foundation models have advanced scene reconstruction and embodied intelligence, scaling to entire cities remains an open challenge, primarily due to the lack of city-scale data. To bridge the gap, we introduce WildCity, a real-world multimodal dataset collected by autonomous fleets traversing compl…
Explore how AlphaEvolve's Gemini-powered algorithms are driving impact across business, infrastructure, and science.
Inference-time scaling for text-to-image generation has progressed from simple Best-of-N (BoN) sampling to guided search methods that verify and steer candidate trajectories at intermediate denoising steps. These approaches focus on when and how often to verify during denoising but largely treat the cost of generation itself as fixed. Moreover, the standard practice of comparing methods by number of function evaluations (NFEs) counts only denoising forward passes and ignores verifier overhead, …
Pixel-wise Earth-observation (EO) foundation models are now achieving state-of-the-art performance via generated spatial embeddings. However, how these models scale and how best to spend a pretraining budget remain poorly understood. We present the largest controlled scaling study for EO to date: 395 training runs on 1,024 GH200 superchips within a fixed pixel-wise Barlow Twins family, each evaluated on 15 downstream tasks. We find that pretraining loss barely predicts downstream performance (|…