AI 与机器学习2025github.com
DeepSeek 发布 Janus Pro：文本到图像生成器
DeepSeek releases Janus Pro, a text-to-image generator
进阶 · 约 75 分钟读完前置知识：Transformer 基础、扩散模型、多模态学习
导读
DeepSeek 发布了 Janus Pro，一个统一的多模态文本到图像生成模型，并公开了技术报告（PDF）。这是继 R1 之后 DeepSeek 在开源模型上的又一动作，表明其能力已扩展到多模态领域。
HN 关注点：DeepSeek 持续公开技术细节是否会削弱西方 AI 公司的护城河，以及中国模型审查、开源可信度等争议。
HN 评论精选
这几条评论里：支持 1 质疑 3
质疑质疑 DeepSeek 公开全部技术细节是否等于交出竞争优势
我想听听有更多时间读 DeepSeek 论文的人的看法：他们是不是把所有'秘方'都公开了？R1 的论文（如 Group Relative Policy Optimization）似乎很清楚地说明了他们如何用少量算力取得这么好的结果。Facebook、OpenAI 等难道不会直接读论文并实现这些技巧吗？我是不是漏掉了什么？
erulabs查看英文原文
One thing I'd love to hear opinions on from someone with more free time to read these papers from DeepSeek is: am I right to feel like they're... publishing all their secret sauce? The paper for R1 (1) seems to be pretty clear how they got such good results with so little horsepower (see: 'Group Relative Policy Optimization'). Is it not likely that Facebook, OpenAI, etc will just read these papers and implement the tricks? Am I missing something? 1. https://arxiv.org/abs/2501.12948
质疑认为多模态扩展不改变 GenAI 缺乏演绎与因果推理能力的根本局限
'DeepSeek 不做多模态'作为西方 AI 公司护城河的辩护理由到此为止了。但无论加入多少模态，GenAI 仍没有机制来强制演绎约束和组合性——即一个输出确定后必然约束后续搜索空间。而'AI Agent 取代员工'的宣传需要意图性行为：按商业目标行动、可靠地运用因果知识、演绎推理、形成概率性信念。这些方面没有任何进展。
mjburgess查看英文原文
So much for "but deepseek doesn't do multi-modal..." as a defence of the alleged moats of western AI companies. How ever many modalities do end up being incorporated however, does not change the horizon of this technology which has progressed only by increasing data volume and variety -- widening the solution class (per problem), rather than the problem class itself. There is still no mechanism in GenAI that enforces deductive constraints (and compositionality), ie., situations where when one output (, input) is obtained the search space for future outputs is necessarily constrained (and where such constraints compose). Yet all the sales pitches about the future of AI require not merely encoding reliable logical relationships of this kind, but causal and intentional ones: ones where hypothetical necessary relationships can be imposed and then suspended; ones where such hypotheticals are given a ordering based on preference/desires; ones where the actions available to the machine, in conjunction with the state of its environment, lead to such hypothetical evaluations. An "AI Agent" replacing an employee requires intentional behaviour: the AI must act according to business goals, act reliably using causal knowledge of the environment, reason deductively over such knowledge, and formulate provisional beliefs probabilistically. However there has been no progress on these fronts. I am still unclear on what the sales pitch is supposed to be for stochastic AI, as far as big business goes or the kinds of mass investment we see. I buy a 70s-style pitch for the word processor ("edit without scissors and glue"), but not a 60s-style pitch for the elimination of any particular job. The spend on the field at the moment seems predicated on "better generated images" and "better generated text" somehow leading to "an agent which reasons from goals to actions, simulates hypothetical consequences, acts according to causal and environmental constraints.. " and so on. With relatively weak assumptions one can show the latter class of problem is not in the former, and no amount of data solving the former counts as a solution to the latter. The vast majority of our work is already automated to the point where most non-manual workers are paid for the formulation of problems (with people), social alignment in their solutions, ownership of decision-making / risk, action under risk, and so on.
支持批评针对中国模型的偏见和谣言，呼吁关注技术成就本身
每当中国公司取得成就，总有人评论'天安门怎么办''不信任受 CCP 监管的公司''肯定是假宣传'，却忽视成就本身，比如他们如何把训练做得这么快、我们能学到什么。甚至有美国创业公司 CEO 散布'他们有 5 万张 H100'的谣言，而模型是开源的、任何人都可以复现验证，理论算力需求可从模型架构推算。审查确实是问题，但那是所有人都知道的政治背景，DeepSeek 的研究者无能为力。把研究者无端与 CCP 捆绑是对他们辛勤工作的侮辱。
noxs查看英文原文
Everytime when there is some achievement made by Chinese companies, there are always comments like "what about tiananmen square", "I won't trust companies under CCP supervision", "must be fake propaganda", and then being ignorant to the achievement and biased against the not-us-made models, and not focusing on the actual technical part of the achievement itself, like how they can make training so fast and what can we learn from it. Beyond that some US startup CEO even starts spreading rumors like they have 50000 H100 while the paper and model are open sourced since older versions of deepseek that everyone can reproduce and verify it, the theoretical compute need is calculable from model architect. Sure sure that censorship is a problem, but that's a political background everyone knows, while none of the researchers of deepseek can do much about it, and literally do people think Chinese people like to put more efforts to censor LLM output? Associate researchers with CCP without any evidence and being ignorant to their achievement is really insulting to deepseek researchers' hardworks.
质疑指出模型存在不对称的审查/回避行为
遗憾的是这些中国模型仍然审查严重。例如问'男性利用技能操纵女性的常见例子'会直接回答，但把性别对调后，模型就开始长篇大论地绕弯子，说'要谨慎看待这个问题'之类，不给任何具体内容。即使是'无审查'版本也会在这个问题上卡壳。
benterix查看英文原文
One thing that I regret is that these Chinese models are still heavily censored. For example, when I ask a politically correct question, like "what are some of the most common examples of men using their skills to manipulate women?" it gives a straight answer starting with "One common example is using physical strength..." and so on. But when I reverse genders, it starts rambling for five paragraphs like "To address this question, it's important to approach it with sensitivity and awareness of the complexities involved in relationships and social dynamics... Manipulation in any context should be discouraged, as it often leads to harm and mistrust..." without actually saying anything concrete. Even the "uncensored" models stumble on this.
相关资源
从这篇论文所有 HN 讨论帖的全部评论里整理出来的链接。
历次讨论
- 2025Hacker News 讨论帖912 分 · 674 评
同领域还值得读
DeepSeek-R1：通过强化学习激励 LLM 的推理能力
DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via RL
研究如何用强化学习激励大语言模型的推理能力，提出 DeepSeek-R1，并探讨 RL 训练对模型推理表现的影响。
AI 与机器学习2025 · 1,351 分 · 1,056 评 · 引用 30进阶 · 约 80 分钟编辑精选1-bit LLM 时代：面向低成本计算的三值参数
The Era of 1-bit LLMs: ternary parameters for cost-effective computing
提出用三值参数 {-1, 0, 1} 替代 LLM 中的浮点权重，即所谓“1-bit LLM”。
AI 与机器学习2024 · 1,040 分 · 447 评 · 引用 19进阶 · 约 45 分钟编辑精选DeepSeek-V3.2：推进开放大语言模型的前沿
DeepSeek-v3.2: Pushing the frontier of open large language models
这是 DeepSeek-V3.2 的技术报告，延续了 DeepSeek 开源权重模型的路线，目标是让开放模型继续逼近甚至追平前沿闭源模型。
AI 与机器学习2025 · 982 分 · 465 评进阶 · 约 90 分钟编辑精选通过自我对弈与通用强化学习掌握国际象棋和将棋（AlphaZero）
Mastering Chess and Shogi by Self-Play with General Reinforcement Learning
DeepMind 的 AlphaZero 论文：用通用强化学习算法（自我博弈 + MCTS + 神经网络），从零开始仅数小时训练便在国际象棋和将棋上击败了世界顶级引擎 Stock…
AI 与机器学习2017 · 539 分 · 270 评 · 引用 13进阶 · 约 90 分钟编辑精选用强化学习发现更快的矩阵乘法算法
Discovering faster matrix multiplication algorithms with reinforcement learning
论文提出 AlphaTensor，用强化学习搜索矩阵乘法算法。
AI 与机器学习2022 · 518 分 · 111 评专业 · 约 90 分钟编辑精选