Junfan Huang

Not a boring researcher with research experience @Cambridge_Uni and @EdinburghUni, now building

[Junfan Huang @junfanhuang]

@thsottiaux Waking up after going /fast be like 💔

[Junfan Huang @junfanhuang]

@Liwei23Wang @aigclink 汉化版可能只需350RMB

[Junfan Huang @junfanhuang]

@RealYDT 一个商业模式是ToC
一个是ToB

[Junfan Huang @junfanhuang]

@hngshng73249148 @Phoenixyin13 这个年代诱惑太多了不是

[Junfan Huang @junfanhuang]

@L0vetodream 那谁来预测一下MacBook pro M6什么时候出

[Junfan Huang @junfanhuang]

@vadiamit How about word level time alignment

[Junfan Huang @junfanhuang]

@GoogleAIStudio How about word level time alignment?

[Junfan Huang @junfanhuang]

@akokoi1 @thsottiaux 35岁好像？

[Junfan Huang @junfanhuang]

@mylifcc @ZpGhxv 推特显示的时间*

[Junfan Huang @junfanhuang]

@mylifcc @ZpGhxv 估计是算错了 11:30发的而我们大概率看到的是第二天了，有两种情况
1. 推特说的时间不准确，往前推几小时还是落在第二天
2. 推特说的准确，而AI没有推理发布时间

[Junfan Huang @junfanhuang]

@ZpGhxv @mylifcc ChatGPT也不能感知成功

[Junfan Huang @junfanhuang]

@zzxwill @noobnooc 或许可以存在codex里面

[Percy Liang @percyliang]

🚢 Marin 535B-A23B started training this week! As usual, the whole process is open.
Voyage plan: pretraining (80%) + midtraining (20%) on 18.75T tokens on 11 x GB200 NVL72 for ~3 months (2.7e24 FLOPs). Post-training will follow.
Before kicking off the run, we trained a 4-rung scaling ladder from 1.6B-A61M (48B tokens) to 27.7B-A1.2B (926B tokens) to debug issues, and to make a forecast of our hero run. This is by far our biggest run, so definitely expecting the unexpected.

[Junfan Huang @junfanhuang]

@Lonely__MH 还可以

[Junfan Huang @junfanhuang]

@karminski3 求一个参数配置，为什么感觉tool use parser和思考模型不太兼容

[Junfan Huang @junfanhuang]

@BohuTANG @ssi 但是他们说 zero data retention

[Junfan Huang @junfanhuang]

我的Qwen 3.8 27B似乎没有听说的那么好用，是我哪里配置得不好么

[Junfan Huang @junfanhuang]

@RealYDT 总不能是SSI的模型吧

[Junfan Huang @junfanhuang]

@editxshub @theo SSI？

[Junfan Huang @junfanhuang]

@theo it’s time to enhance your vibe coding skills (just kidding), get well soon