Pinned Tweet
What each budget can buy?
The Pareto front of local AI right now:
modeldock.net/insights/local…
Nothing demoralizing a tech/engineering company more than appointing a financial guy as the CEO.
Declare to public that the board has no clue the company does.
2
4
493
My prediction is that, within 18 months, the entire AI supply chain/with white house/s help, will be back begging China to buy more chips, with data center constructions in US essentially dead on arrival and previous committed buyers goes default.
So many times in history.
600,000 Nvidia chips are stuck in Hong Kong because Chinese customs officials are refusing to let them enter the market, to avoid dependence on the American brand, and allow local firms to come up with their own solutions.
That is technological sovereignty at work.
And the new storyline today is that Jensen Huang is trying to circumvent the sanctions by secretly selling chips to Megaspeed, where his friend Alice Huang works, so she can sell them to China ?
We’ve gone from “Chinese spy” to "mistress" to “executive of an Nvidia client company circumventing sanctions” in 24 hours, but none of this holds water.
Pushing Huang out, taking control of his company because it had the world’s largest market capitalization for a year, and the fact that Huang doesn’t take his orders from the White House or Silicon Valley, and has a different opinion on how to engage China : these are the “crimes” for which he is being attacked.
Sitting on the fence is not permitted.
自从Claude Opus5.5发布以来,
明显感到M,G这几家公司的Agent开始变聪明了,模型研发开始更顺利了,产品改进频率变得更快了。
RSI,但是并不一定需要是本公司的LLM。
Only thing that is a bit concerning is $NVDA now losing more bargaining power to CSPs now SEA export becomes extra tight.
1
2,139
毕竟一个成年女性,有千丝万缕的社会关系和难以捉摸的背景:
有可能是竞争对手派来的,
有可能是来分家产的,
有可能是老婆闺蜜,
竟然还有可能是个间谍,
万一要是朝鲜间谍,就万劫不复了,
富豪们想了想,禁不住轻抚钱包里爱泼斯坦的遗照。
1
15
3,839
马达尼当上纽约市长之前的共和党:
"他要搞社会主义,他会让纽约万劫不复。"
马达尼当上纽约几个月后:
1. 万斯出来领功,表示是特朗普政府让纽约变得更好(屁)
2.普通人视角:这个社会主义,搞普惠日托,搞保障房建设,搞市政建设提供工作,控制犯罪,这就是你们共和党一直批评的社会主义?
3.共和党视角:他居然给穷人福利,用白花花的银子让穷人吃饱穿暖,有地方住,银子怎么能用来造保障房而不是数据中心,造孽啊。
美国的国民性格里,最特别的一点,就是自大。
明明全国也建不了多少数据中心,但是满脑子想的都是怎么卡中国的脖子。
老登政治家这么想也情有可愿,年轻一辈的AI公司也是这个调门。
吃不下,可是真的会噎死的。
一段往事,美国20年代对苏联禁运,30年代陷入经济危机了把工厂整个往苏联卖。
如果我们客观的去考虑,
OAI如果能保持今天A社2/3的营收就很好了,
那么OAI的算力开支也应该是A社的2/3左右,即4000亿左右。
四千亿也已经很多了,但是还是远远小于OAI这些疯子承诺的1.4万亿。
也就是,现在至少有一万亿的overbuild, 即使不考虑 deepinfer / Strata 带来的生产力暴增。
不知道这是哪个幸运儿负责这一万亿建设。
OpenAI和Anthropic两家公司,公开披露的长期算力及基础设施承诺,合计已达约2万亿美元量级。
你们认为,这2万亿美元最终能兑现多少?
OpenAI约1.4万亿美元:据Reuters报道,Altman称公司承诺未来八年投入这一金额建设算力资源。
Anthropic约6370亿美元:根据其招股文件的梳理,包含已签协议及预计交易。
2
15
8,355
Big if true..
这位大神宣称可以用更智能的LLM去读取/理解/修改LLM的权重,从而跳过训练,直接让低阶LLM更聪明。
我看了看手边的Opus5.5 和Qwen Flash Next。
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…


























