AI早报 · 2026年2月5日

以下为2026年2月5日（昨日）AI领域热点速览，覆盖模型/产品发布、研究进展、开源生态与产业动态。内容来自各机构官方博客与arXiv订阅源。

20条AI热点

1. GPT-5 lowers the cost of cell-free protein synthesis

   An autonomous lab combining OpenAI’s GPT-5 with Ginkgo Bioworks’ cloud automation cut cell-free protein synthesis costs by 40% through closed-loop exp。 来源：OpenAI

2. Navigating health questions with ChatGPT

   A family shares how ChatGPT helped them prepare for critical cancer treatment decisions for their son alongside expert guidance from his doctors.。 来源：OpenAI

3. GPT-5.3-Codex System Card

   GPT‑5.3-Codex is the most capable agentic coding model to date, combining the frontier coding performance of GPT‑5.2-Codex with the reasoning and prof。 来源：OpenAI

4. Introducing GPT-5.3-Codex

   GPT-5.3-Codex is a Codex-native agent that pairs frontier coding performance with general reasoning to support long-horizon, real-world technical work。 来源：OpenAI

5. Introducing OpenAI Frontier

   OpenAI Frontier is an enterprise platform for building, deploying, and managing AI agents with shared context, onboarding, permissions, and governance。 来源：OpenAI

6. Introducing Trusted Access for Cyber

   OpenAI introduces Trusted Access for Cyber, a trust-based framework that expands access to frontier cyber capabilities while strengthening safeguards。 来源：OpenAI

7. Watch our new Gemini ad ahead of football’s biggest weekend

   Learn more about Google’s new ad that will run during football’s Big Game on February 8.该信息来自Google AI Blog，建议结合原文核对关键指标、发布时间与适用范围。 来源：Google AI Blog

8. Paza: Introducing automatic speech recognition benchmarks and models for low resource languages

   Microsoft Research unveils Paza, a human-centered speech pipeline, and PazaBench, the first leaderboard for low-resource languages. It covers 39 Afric。 来源：Microsoft Research

9. Alignment Drift in Multimodal LLMs: A Two-Phase, Longitudinal Evaluation of Harm Across Eight Model Releases

   arXiv:2602.04739v1 Announce Type: cross Abstract: Multimodal large language models (MLLMs) are increasingly deployed in real-world systems, yet their。 来源：arXiv cs.AI

10. RASA: Routing-Aware Safety Alignment for Mixture-of-Experts Models

    arXiv:2602.04448v1 Announce Type: new Abstract: Mixture-of-Experts (MoE) language models introduce unique challenges for safety alignment due to their。 来源：arXiv cs.LG

11. RASA: Routing-Aware Safety Alignment for Mixture-of-Experts Models

    arXiv:2602.04448v1 Announce Type: cross Abstract: Mixture-of-Experts (MoE) language models introduce unique challenges for safety alignment due to the。 来源：arXiv cs.AI

12. The latest AI news we announced in January

    Google AI announcements from January该信息来自Google AI Blog，建议结合原文核对关键指标、发布时间与适用范围。如果你在做决策，可重点关注：方法/数据来源、边界条件、成本与潜在风险。 来源：Google AI Blog

13. WIND: Weather Inverse Diffusion for Zero-Shot Atmospheric Modeling

    arXiv:2602.03924v1 Announce Type: new Abstract: Deep learning has revolutionized weather and climate modeling, yet the current landscape remains fragm。 来源：arXiv cs.LG

14. A Probabilistic Framework for Solving High-Frequency Helmholtz Equations via Diffusion Models

    arXiv:2602.04082v1 Announce Type: new Abstract: Deterministic neural operators perform well on many PDEs but can struggle with the approximation of hi。 来源：arXiv cs.LG

15. CoRe: Context-Robust Remasking for Diffusion Language Models

    arXiv:2602.04096v1 Announce Type: new Abstract: Standard decoding in Masked Diffusion Models (MDMs) is hindered by context rigidity: tokens are retain。 来源：arXiv cs.LG

16. Theory of Speciation Transitions in Diffusion Models with General Class Structure

    arXiv:2602.04404v1 Announce Type: new Abstract: Diffusion Models generate data by reversing a stochastic diffusion process, progressively transforming。 来源：arXiv cs.LG

17. Rethinking the Design Space of Reinforcement Learning for Diffusion Models: On the Importance of Likelihood Estimation Beyond Loss Design

    arXiv:2602.04663v1 Announce Type: new Abstract: Reinforcement learning has been widely applied to diffusion and flow models for visual tasks such as t。 来源：arXiv cs.LG

18. Dynamical Regimes of Multimodal Diffusion Models

    arXiv:2602.04780v1 Announce Type: new Abstract: Diffusion based generative models have achieved unprecedented fidelity in synthesizing high dimensiona。 来源：arXiv cs.LG

19. Sparse-to-Sparse Training of Diffusion Models

    arXiv:2504.21380v2 Announce Type: replace Abstract: Diffusion models (DMs) are a powerful type of generative models that have achieved state-of-the-ar。 来源：arXiv cs.LG

20. Discrete Diffusion-Based Model-Level Explanation of Heterogeneous GNNs with Node Features

    arXiv:2508.08458v2 Announce Type: replace Abstract: Many real-world datasets, such as citation networks, social networks, and molecular structures, ar。 来源：arXiv cs.LG

趋势点评：从昨日信息密度来看，官方渠道更倾向于用“产品化能力+生态工具链”来讲述AI进展，而研究侧则继续在效率、对齐与多模态数据构建上加速迭代。接下来一周值得重点跟踪：是否出现更明确的企业级落地指标（成本、时延、可靠性）以及开源社区对新范式（Agent、长上下文、工具调用）的吸收速度。