Jin

tech lead and research scientist, calmly computing @MegaETH. All opinions are my own.

[Jin @jinfizz]

Many folks were suprised by how fast I decided to leave my PhD program at Harvard and join MegaETH two months ago. To be honest it was such an easy choice. Any systems researcher would do the same after reading @yilongl_megaeth's analysis of Etheruem performance and chatting with @yangl1996.

[Jin @jinfizz]

antigravity is the only agent app that stuck at google login redirection

[Jin @jinfizz]

@yifanxu_ephai 上学的时候实验室都用slack无一例外 课程一般用专门的课程平台

[Jin @jinfizz]

@implied_vol I've saw the same disclosures. still want the 8x speed tho

[Jin @jinfizz]

when you pay $500 a month only to try ultrafast

good job openai, really good

[Jin @jinfizz]

@yaogangqiang 这文章大概率不是想给你看的 是给移民局看的

[Jin @jinfizz]

Been running deepswe with deepseek v4.1 flash for fun. Interestingly it recognizes deepswe question format and always tries to guess what hidden tests would be instead of programming as an expert. In its thinkings, keyword "hidden tests" just comes out of nowhere. RL I guess.

[Jin @jinfizz]

@maria_rcks seriously it's so dumb. openai is just playing with us

[Jin @jinfizz]

@AIandDesign @OpenAI @thsottiaux https://t.co/HB2Q85BcT4

[Jin @jinfizz]

根据我几天的实测，codex会在拥挤时自动把astra降级成5.6 luna，有的账号会直接全局默认降级，没有任何通知，且应该按照astra价格收费。

如果你也有一样的担忧，可以用我写的监控器，默认半小时监控一次。

https://t.co/cDKVo1D9Fc

[Jin @jinfizz]

@MmuzTW @OpenAI @thsottiaux @sama yo dude, why are you mentioning Chinese? Chinese AI community was always among the first on X pointing out the scam (gpt got nerfed or quota got decreased) and delievered ways to detect it

[Jin @jinfizz]

因为gpt降智被迫和fable讨论一个项目，聊了两天感觉智商有大问题，干活也不认真讲细节，各种擅自作主的错误，导致实验结论都不可信，血压蹭蹭涨

[Jin @jinfizz]

@servasyy_ai 这根本不对。jev无法动态问这些问题或者填充答案，除非带个其它模型。而如果带了其它模型，也就用不到jev。

[Jin @jinfizz]

@synthwavedd https://t.co/P5VyahL6bb

[Jin @jinfizz]

根据我几天的实测，codex会在拥挤时自动把astra降级成5.6 luna，有的账号会直接全局默认降级，没有任何通知，且应该按照astra价格收费。

如果你也有一样的担忧，可以用我写的监控器，默认半小时监控一次。

https://t.co/cDKVo1D9Fc

[Jin @jinfizz]

@libukai @attaalong 我觉得可以在准备数据的时候尽量把有一定关联度的东西一并附上（比如提前建立一张图，把相关的1-hop节点都给上）

[Jin @jinfizz]

@masteralanlab 让claude分析了一下似乎并不是这个原因。我本地降智的记录和这个292对不上。而且不存在这个状态码

[Jin @jinfizz]

根据我几天的实测，codex会在拥挤时自动把astra降级成5.6 luna，有的账号会直接全局默认降级，没有任何通知，且应该按照astra价格收费。

如果你也有一样的担忧，可以用我写的监控器，默认半小时监控一次。

https://t.co/cDKVo1D9Fc

[Jin @jinfizz]

@zhangsen125899 也是一样可以比较，但是pro版好像暂时没有fingerprint

[Jin @jinfizz]

@djn208097 哈哈，可以让agent改造一下cli

[Jin @jinfizz]

@VoiceOver0000 linux可以用cli

[Jin @jinfizz]

@Tokemacc monitoring is essential in this case. I can avoid continuing the work, so that it's not polluted by small models

[Jin @jinfizz]

@eDust2014 why not? It's basically comparing fingerprints of models with extremely high confidence

[Jin @jinfizz]

@qldTowers yeah many people say it's based on IP