腾讯发布了Tencent HY4预览版并开源,这是一个23x23像素的小图像,可能是标志、图标或头像。
正文
根据提供的文章内容,主要信息如下:腾讯发布了并开源了Tencent Hy4预览版。文章标题为“X-1.svg”,来源链接指向腾讯官网,发布时间为2024年5月17日。页面内容包含一个23x23像素的小图像,可能是一个标志、图标或头像。
评论
根据评论内容,总结如下:
主要观点与论据:
性能与成本优势(评分:None,作者:minimaxir):Hy4在OpenRouter上处理速度极快,数天内处理数万亿token,远超GLM 5.3一周的量。其缓存成本仅5%,低于行业10%-20%,更具吸引力。
- 关键引用:"Hy4 apparently has ludicrous traction on OpenRouter already... with trillions of tokens processed in a couple days"
- "it's relatively cheap with a 5% cache cost when everyone is still doing 10%/20% cache costs"
速度与可用性担忧(评分:None,作者:vcryan):对Hy3的速度不满,尽管模型较小,但推理速度慢,包括腾讯直接提供的服务。期待Hy4改进。
- 关键引用:"My one concern over Hy3 was speed... it was relatively slow everywhere I could get it"
- "it could be served much faster as a smaller model"
模型质量认可(评分:None,作者:jorl17):Hy3在通用代理任务中表现优异,仅次于DeepSeek4-Flash,接近其行为,怀疑可能基于DeepSeek分支。
- 关键引用:"as a general purpose agentic model, it was only beaten by deepseek4-flash in our tests"
- "it was so close to deepseek behaviour I kept thinking it must have been forked from it"
开源争议(评分:None,作者:petcat):批评“开源”说法,认为仅开放权重而非真正开源,模型是“不透明的二进制块”,类比Photoshop二进制文件。
- 关键引用:"Open weight is not the same as open source... just opaque binary blobs you can run on your own computer"
- "Imagine thinking that running a Photoshop binary on your own computer... means that it's 'open source'"
图表与展示问题(评分:None,作者:usernomdeguerre、fastball):指出博客中柱状图高度与数值不符,以及表格高亮方式不当,呼吁停止“图表犯罪”。
- 关键引用:"Higher numbers don't seem to correspond correctly to their actual height?"
- "if you're gonna highlight a winner in a table of benchmarks, don't highlight your entire model row"
技术进展与地缘政治(评分:None,作者:codethief):认为Hy4的递归自我改进循环预示中国AI已赶上,无需窃取技术,反驳“中国落后”预测。
- 关键引用:"No need to steal anything, they have already caught up"
- "This established an early-stage recursive self-improvement loop"
其他评论:部分用户表达支持(如“Go go China!”),或认为LLM在繁琐任务上提升生产力,但缺乏品味;也有用户因超时/限速无法测试。
平衡性总结:
评论呈现两极分化:一方面认可Hy4的性能、成本优势及中国AI进步,另一方面批评速度、开源定义、图表展示及可用性问题。整体上,正面评价集中于技术指标,负面评价涉及透明度与用户体验。