原贴链接

帖子内容仅为一个图片链接，无实质可翻译内容

讨论总结

本次讨论围绕Deep Seek v3展开，涵盖了它的多个功能方面，如“Deep Think”选项的创新性、R1 - lite - preview版本的相关信息、搜索按钮功能、是否有深色模式、API情况等，既有对新功能的好奇与着迷，也有对部分功能是否首创的质疑，用户之间进行了信息交流与观点分享。

主要观点

  1. 👍 对Deep Seek v3中“Deep Think”选项的创新性表示怀疑
    • 支持理由：怀疑只是切换模型至R1，没有更多创新之处
    • 反对声音：有人对“Deep Think”选项着迷，认为像是窥探编程天才的思维
  2. 🔥 Deep Seek v3的R1 - lite - preview版本在编码任务上比ChatGPT更适合非编码人员
    • 正方观点：在编码需求方面比ChatGPT好，脚本错误少、理解指令等
    • 反方观点：无（未提及）
  3. 💡 OpenAI最早于9月o1系列初始发布时就有类似“Deep Think”功能
    • 解释：回复者DFructonucleotide指出OpenAI才是最早有该功能的

金句与有趣评论

  1. “😂 Is that not just switching the model to R1?”
    • 亮点：直接对“Deep Think”选项的本质提出疑问
  2. “🤔 I’ve been playing with this version over last week, and I find it was very good for my coding needs. For me, it is better than anything OAI offers under ChatGPT for coding tasks.”
    • 亮点：通过自身使用体验对比Deep Seek v3和ChatGPT在编码任务上的优劣
  3. “👀 More interested in the new search button. Similar to perplexity but free…”
    • 亮点：表明对Deep Seek v3搜索按钮功能的兴趣以及其免费且与Perplexity类似的特性

情感分析

总体情感倾向较为中性，主要分歧点在于对Deep Seek v3新功能的评价，部分人怀疑新功能创新性不足，而另一部分人则对新功能表现出着迷和认可。可能的原因是用户对功能的期望和使用场景不同，以及对技术的理解程度有差异。

趋势与预测

  • 新兴话题：对Deep Seek v3的本地使用可能性以及未来开源权重后的影响。
  • 潜在影响：如果Deep Seek v3的R1 - lite - preview版本开源权重，可能会对本地模型的发展产生影响，吸引更多人使用并开发基于该版本的应用。

详细内容：

标题：Deep Seek v3 的“Deep Think”选项引发热议

近日，Reddit 上一则关于 Deep Seek v3 具有“Deep Think”选项的帖子引发了众多网友的关注。该帖子获得了大量的点赞和评论。帖子中提到了这个选项，并附上了相关链接https://chat.deepseek.com/。主要的讨论方向包括该选项是否只是切换到了 R1 模型、其在编码任务中的表现、是否会有 API 开放、与其他类似产品的比较等。文章将要探讨的核心问题是 Deep Seek v3 的“Deep Think”选项的创新性和实用性。

在讨论中，有人认为这可能只是切换到了 R1 模型。比如有人说：“[Few_Painter_5588] Is that not just switching the model to R1?” 但也有人提出不同看法。

有用户分享道：“[DFructonucleotide] That is R1 -lite -preview. It has been online since Nov 20, likely based on some smaller deepseek v2 variant. No API currently but they said it will be released open weight when it’s fully ready. Also the v3 technical paper said that an internal version of R1 has been extensively utilized for post training of v3, which explains some of its math and coding capabilities. Fascinating indeed.”

还有用户表示：“[dodo13333] I’ve been playing with this version over last week, and I find it was very good for my coding needs. For me, it is better than anything OAI offers under ChatGPT for coding tasks. I’m no -coder. It produced scripts with fewer errors, model had a better understanding of my instructions, and overall experience was great. It even managed to split single script to 5 smaller scripts and worked with them coherently afterwards. That also improved its success rate. I wish we could get this one for local use…”

有人认为如果该选项是基于 27B MoE in deepseek vl2，激活参数约 4B，那么对于笔记本电脑甚至使用 CPU 进行推理将是理想的。比如：“[DFructonucleotide] I would like to have it too! It is very likely based on the 27B MoE in deepseek vl2, with around 4B activated params. That would be ideal for a laptop, even using cpu for inference. If true its coding and math capabilities would completely destroy all local models.”

对于该选项的实用性，有人表示非常痴迷，就像用户[SentientBaba420]说的：“Yooo, I’m literally obsessed with Deep Seek v3’s ‘Deep Think’ option rn. Like OP said, it’s like having a glimpse into the mind of a coding genius.”

讨论中的共识在于大家都对这个新选项充满了好奇和期待。特别有见地的观点是关于其技术原理和潜在应用的分析，丰富了对该选项的理解。

总之，关于 Deep Seek v3 的“Deep Think”选项的讨论展示了大家对新技术的关注和探索，未来其发展和应用值得持续关注。