Zephyr
@zephyr_z9
Followers
7K
Following
50K
Media
1K
Statuses
17K
My Third Post.Implications of the H20.It will have profound change on the dynamics of US-China AI race especially in the RL/inference age
3
4
63
RT @pandawatch88: @EleanorOlcott @zijing_wu and God bless them because Kling is far faaaar better than VEO in picture to video generation.
0
1
0
What is the moral of the story here, anon??.
>Wife leaves.>Poasting game instantly improves.
4
0
11
>Wife leaves.>Poasting game instantly improves.
Most research published in Science or Nature journals are not trustable
7
6
333
RT @zijing_wu: Glad this is finally out after months of old-school journalism. There’s so much more to the story - and it needs to be told.….
ft.com
Roaring black market for US semiconductors operates despite efforts to curb Beijing’s high-tech ambitions
0
13
0
RT @tphuang: We have finally reached the era of Robocop in real life. Well, at least in Shanghai. Technically, it is still in training. htt….
0
2
0
Even 3000-pound machines are getting smuggled.(Though I shouldn't be surprised, considering the volume of drugs & arms that get sold illegally).
Crazy headline but. 1,000,000,000 ("1bn")÷489,000/racks (8xB200) = 16360 (2045 x 8) B200 GPUs.
1
0
3
It has gone down.Few months ago, they were selling for over $600K while SuperMicro/Dell was selling it for $450K in US.
The FT is claiming that more than $1 billion worth of NVIDIA chips have made their way into China over the last three months, including Blackwells which they claim are 'widely available.' They report that the current market price for a rack of eight B200's is $489,000.
1
0
20
Bruh.Just one company has shipped over 8000 B200s (20K H200 equivalents) in the last 3 months in China.Egregious
3
1
13
A 355B should be quite good.I think they will surpass the latest models from Qwen
@zephyr_z9 You were right it's GLM
2
0
16
they can't even catch a 3000-pound shipment.
Wait, is this for real? I was well aware of B200s being smuggled into China, but GB200s too? Has that actually happened?
0
0
3
Scalping will also increase a lot due to AI agents.
Will AI agents mean that hard-to-reserve restaurants book up immediately once dates become available? One of many AI implications that trouble me.
3
0
12
Why are they asking bullshit like this??.
The design process of HLE required the questions to be unanswerable by contemporary LLMs. That lead to many gotcha style questions like the one below. It’s a trick question – in 2002, a few atoms of a group 18 element Oganesson were made for a few milliseconds. 2/7
2
1
10
this is very cool.
Imagine generating an explorable world from a single image. Meet Yume: a system that turns images into dynamic, controllable 3D video worlds—navigated with keyboard. For this, they introduce a well-designed framework, which consists of four main components, including . - camera
0
0
2
need big breakthroughs in multimodality.
CEO of Google Deepmind Demis when asked about AGI:. “50% chance by 2030” . I agree with his definition for AGI as well — being as cognitively capable as any human in any task
0
0
9
the nine tailed fox has been exorcised.
Hello everyone,. I wish to share that I have now relocated my base of operations to New York and, for a variety of both scientific and personal reasons, Jiankui and I will be pursuing separate paths. My professional and personal commitments are now rooted in North America, while.
3
0
44
RT @NateHindenburg: @TuckerCarlson I am Nate Anderson, the founder of Hindenburg Research referenced repeatedly in this bizarre and fantast….
0
1K
0
Zhipu was going to open source their 100B model.It's likely them.
2
0
26
Why is ScaleAI publishing the stuff closed Labs use to train their reasoners??.
Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains. 'We introduce Rubrics as Rewards (RaR), a framework that uses structured, checklist-style rubrics as interpretable reward signals for on-policy training with GRPO. Our best RaR method yields up to a relative
0
0
5
Here we go again.
Another massive open-source LLM is coming soon… and it's from a Chinese company too.
7
5
113
This graph is wrong if u consider the amount of fab capacity coming online next year and Nvidia lacking any good training chips for China.
0
0
5