xerwanderer @xerwanderer

He who desires but acts not, breeds pestilence. | INFP Shanghai Joined July 2010

Tweets

7K
Followers

634
Following

2K
Likes

18K

*Walter Bloomberg @DeItaone

2 months ago

TRUMP: EVERYBODY PAYS ON AUGUST 1ST

23 21 222 58K 2

xerwanderer @xerwanderer

4 months ago

Intelligence in, intelligence out. An AI mirrors how we talk, think, and act. btw 上次读到类似感受的文字是 @sama preview upcoming creative writing model，所以啥时候才能看到

0 0 1 167 1

Download Image

xerwanderer @xerwanderer

4 months ago

Meta AI 中文译名：“元宝”

0 0 0 145 0

Congrats OpenAI: Game changer. More and more benchmarks are being saturated. And it is becoming increasingly difficult to tell how good the models really are. They are now so intelligent (I am aware of the problems with the term) that it is difficult to say unequivocally how…

44 62 675 70K 102

Download Image

xerwanderer @xerwanderer

5 months ago

啥时候上限啥时候续订 Pro（

Sam Altman @sama

5 months ago

啥时候上限啥时候续订 Pro（

462 373 8K 1.3M 374

0 0 0 156 0

Shunyu Yao @ShunyuYao12

5 months ago

I finally wrote another blogpost: ysymyth.github.io/The-Second-Hal… AI just keeps getting better over time, but NOW is a special moment that i call “the halftime”. Before it, training > eval. After it, eval > training. The reason: RL finally works. Lmk ur feedback so I’ll polish it.

36 198 1K 282K 751

xerwanderer @xerwanderer

5 months ago

为什么 OAI 新模型的 knowledge cutoff 还是 24 年一月？

0 0 0 73 0

xerwanderer @xerwanderer

5 months ago

Moonshot: CoT reward model 极大的提升了准确率 DeepSeek: 那让我们在 reward model 上也做 RL 吧（

0 0 2 94 0

xerwanderer @xerwanderer

5 months ago

Notes on PaperBench Evaluating AI’s Ability to Replicate AI Research. （对于没有太多的 multip-hop 任务专门的 post training 的 LLM）做 budget forcing 去显著提高 agent 在复杂任务上的表现