DeepMind

From Generation to Perception: GenCeption’s Architectural Trade-Offs and Empirical Boundaries

by

The research team introduces GenCeption, which turns a generative diffusion model into a feed-forward visual perception tool. On the benchmarks selected in the paper, the architecture achieves performance comparable to task-specific models using roughly 1/7 to 1/500 of the task-training data. This article analyzes its mechanism, experimental evidence, and engineering trade-offs.

Pairwise Preference Reinforcement Learning in Long-Horizon Decisions: The Mechanisms and Boundaries of Markov Decision Contest

by

In long-horizon decision-making, scalar rewards are difficult to define precisely, while pairwise preference methods have gaps in their theoretical guarantees. Markov Decision Contest proves that its exact solution is in P and introduces the HPI and HPI-Clip algorithms. Experiments across seven high-dimensional control tasks show that the approximate HPI-Clip method learns more efficiently than SPPO. The result defines a new set of boundaries and trade-offs for long-horizon decisions based on pairwise preferences.

長程決策中的成對偏好強化學習：Markov Decision Contest 的機制與邊界

by

長程決策中，標量獎勵難以精確定義，而成對偏好方法在理論保證上存在缺口。研究提出的 Markov Decision Contest 模型將求解複雜度證明為 P 類，並提出 HPI 與 HPI-Clip 算法。實測顯示近似算法在多個高維任務中學習效率優於 SPPO，為長程成對偏好決策提供新的邊界與取捨。

Chinchilla: Model/Data Ratios on a Budget

by

When compute is limited, should you buy a bigger model or feed it more data? Chinchilla gave a counterintuitive answer in 2022: at a fixed training budget, a 70B model trained on 1.4T tokens outperformed a 280B model trained on 300B tokens, using the same compute. That points to an approximate 1:20 parameter-to-token ratio in that setup. If you’re planning an LLM training run, this paper offers a practical way to think about model size, data volume, and where your next dollar creates more learning.