Anathemata (@Faenerator) header
Anathemata (@Faenerator) profile picture

Stravaiging. Persiflage. Little original input. Just stuff I find. Retweets are not endorsements. Not looking for followers. Competent Clerk. Αἰδώς.

48°52.6′S 123°23.6′W
Joined June 2013
Photo by Anathemata (@Faenerator)Photo by Anathemata (@Faenerator)Photo by Anathemata (@Faenerator)Photo by Anathemata (@Faenerator)Photo by Anathemata (@Faenerator)Photo by Anathemata (@Faenerator)Photo by Anathemata (@Faenerator)Photo by Anathemata (@Faenerator)Photo by Anathemata (@Faenerator)Photo by Anathemata (@Faenerator)
Anathemata retweeted
天呐,海底光缆统治世界的时代要结束了,又要见证历史了🤯。。 马斯克首次公布 Starlink V4 的极端参数:性能直接跳涨一百倍,光速天然快了一点四倍且不用绕大陆架,上千万条激光链路正在把地球通信主干搬上深空。 绝大多数人至今依然以为,星链只是给偏远山区、游艇和飞机提供网络补丁的边缘工具。 直到马斯克刚刚在最新的投资人会议上摊开底牌,很多人才第一次看懂 SpaceX 真正的阳谋:把构筑人类现代文明的海底光缆网络,从大洋深处连根拔起,彻底搬到太空。 这里藏着一个极度反直觉的物理常识:
光在太空真空中的传播速度,比在玻璃光纤里整整快了 40%。 
更致命的是几何路径的差异。海底光缆必须顺着大陆架和海沟蜿蜒曲折地绕远路,而太空中的卫星激光链路,两点之间永远是沿着大圆航线在真空中直线对穿。这意味着在跨洋超长距离通信上,太空激光的延迟天然比地面光缆更低。 而阻碍这件事过去的唯一瓶颈,是带宽密度。 马斯克这次直接给出了技术迭代的量化终局:
从 V2 到即将落地的 V4 卫星,单星吞吐能力将直接迎来 100 倍的暴力跃升。
在更长期的宏大蓝图里,超过 10 万颗卫星组成的轨道阵列,每颗都将配备大约 100 个传输速率超 10 Tbps 的光学激光器,在地球外围织成一张百 Exabit 级别的无缝高速网状路由。 这已经不是在做卫星宽带了,这是在用火箭把全球互联网的核心骨干网整个重写一遍。 海底光缆容易被地质地震震断、被船锚挂伤,更深陷地缘博弈的物理封锁;而悬在头顶的真空激光网络,既没有折射损耗,也没有无法逾越的地理边界。 人类用了上百年时间,把铜线和光纤一根根沉入海底来连接彼此;而未来的文明动脉,注定要在没有阻力的星空里以光速奔流。
马斯克刚刚把全球半导体制造行业最后一点虚伪的遮羞布撕碎了, 老马透露今年全世界除了他以外生产的所有 AI 芯片加起来大概只有二十吉瓦,而他正在秘密推进的超级芯片工厂 TeraFab,终极目标是把年产量直接干到整整一太瓦,也就是明年全球总产量的三十倍。 当大多数科技巨头还在为了抢台积电几万片先进封装配额打破头的时候, 马斯克对 TeraFab 这个名字给出的三重拆解,已经把工业狂人的冷血算盘展现得淋漓尽致,它既谐音代表让人毛骨悚然的恐怖, 也代表扎根于这颗星球的土地,但底层最纯粹的本质就是物理学上的太瓦,一个大到超出人类现行半导体常识的超级怪兽。 按照目前的全球供应链节奏,明年全世界哪怕把吃奶的力气用上,全部 AI 芯片的总功耗产能预计也就只能爬坡到三十五到四十五吉瓦, 在传统半导体代工厂眼里这已经算得上是跨越式的军备竞赛,但放在马斯克以太瓦为单位丈量的物理世界里,整个人类现存的芯片制造体系加在一起,弱小得就像是手工作坊在小打小闹。 这其实彻底挑明了软硬件协同演进的一条绝情铁律,任何把通用人工智能挂在嘴边、却把晶圆制造和先进产能全部寄托给外部代工的软件公司, 本质上都是把自家大厦建在了极其脆弱的流沙上,只要产能杠杆的阀门依然被传统供应链卡在几十吉瓦的窄门里,大模型参数狂飙的尽头就必然是一堵无法逾越的物理高墙。 别再觉得一太瓦的超级工厂只是他在麦克风前的随口狂想了,当一个狠人把工业制造的目标定在超出全世界三十倍的极限位置上,他从一开始就没打算跟现有的玩家在同一个棋盘里抢饭吃,他要做的,是用工业怪兽的绝对吨位,把人类文明推进下一个计算纪元。
325
856
108
4,645
960,598
“A review of 104 studies found little evidence that demographic diversity makes scientific teams more productive or influential: Only 15-28% of reported results clearly supported the idea, while most found no effect or the opposite.” stevestewartwilliams.com/p/f…
20
206
21
1,058
39,152
Anathemata retweeted
Productivity before teams meetings
21
615
42
11,766
168,042
Anathemata retweeted
Huge FT letter if you studied Ancient Rome and now work on AI (there are dozens of us!) comparing AI stack to ancient grain supply
8
56
5
402
16,831
Anathemata retweeted
'The wind is always blowing somewhere' Here's the realtime chart of winds in Europe. Blue = light/no winds, orange = lots There is pretty much NO wind anywhere in Europe. So nobody is exporting to save our bacon. We need to rely on our own resources Source: Windy.com
8
31
1
89
1,190
Anathemata retweeted
The fake capacity load factors assumed by the British government are a huge scandal. Where's parliament's investigation into this?
Replying to @7Kiwi
The original Whitelee turbines have underperformed. Average load factor ~20.5% (original) and 16.7% (extension). Adjusted for constraints: about 22%. Government models assume 36–48%. Real performance has been far worse. (2/9)
1
8
16
1,402
Almost all the growth in global coal use since the late 1980s has come from China. Outside China, the world’s use of coal has barely changed since the late 1980s. It was just over 20,000 terawatt-hours back then, and almost four decades later, it is still the same. But the change has been very different in China. The country has increased the use of coal enormously, and today, China is burning more coal than the rest of the world combined. The future of global coal use, and the emissions that come with it, depends heavily on what happens in China. (written by @MaxCRoser)
25
152
29
403
23,546
Don’t be fooled by the boring diplomatic language in the leaked UK-EU draft statement. Under Point 65, Brussels is attempting a massive regulatory hijack of the UK’s highest-growth sectors—specifically AI, quantum computing, and advanced semiconductors.
18
287
39
1,027
45,276
Anathemata retweeted
AI will be the first tech capex boom that Britain misses out on in the modern era
21
42
16
302
24,986
Anathemata retweeted
AI bears need to face the facts that most AI capex is not actually debt funded
20
10
4
98
11,124
In other words, the UK would sacrifice its £28B Agtech sector to prioritise its food and live animal exports to the EU, which amounted to £12B in 2025. For the sake of British innovation, we very much hope this isn't the case.
What is now being briefed on the UK-EU SPS agreement is that the UK didn't achieve any carve outs on gene editing. The EU merely agreed the UK could transition towards the EU's more restrictive approach to gene editing which is due to be implemented in July 2028.
3
98
7
266
7,466
Below is a bare bones outline of a potential option when seeking to develop a durable mechanism for regulating migration flows. It was submitted the Home Office and Number 10 back in 2016, and to a selection of politicians since then. Result: it died on arrival in all those places, reflecting the usual unwillingness to even give consideration to anything that would disrupt a conventional unwisdom. In one sentence, the idea is to turn residency rights / ILRs into financial assets which have first to be acquired, but which then have a redemption value when surrendered (e.g. on subsequent exit from the UK). The control variable is the stock of rights in issuance (not the flows). If the stock is fixed (by government), the flows are governed by the constraint that any new acquisition has to be matched by a redemption. Acquisition values are determined by a familiar interplay of supply and demand. Redemption values provide incentives for exit (re-migration). There is plenty of flesh to be added to the bare bones, and various narratives that could be adopted to explain the functioning of the mechanism -- e.g. it's like acquiring and disposing of a land-use lease -- but I've given up working on those, because it is wasted effort in the face of lack of interest in the basic idea. Posted before, but evergreen: "The difficulty lies, not in the new ideas, but in escaping from the old ones." (Keynes)
2
1
5
436
Anathemata retweeted
Labour's new 'founder relief' is basically a scam. The rules are so tight that even most successful fast-growing tech companies wouldn't qualify, reports @CityAM. A claimant must have: - held at least 25% of the ordinary shares and votes when trading, or preparations for it, began - held them for five years, - been a UK resident for five continuous tax years. Their company needs £10m of turnover and 15 full-time staff, and pass two of three tests: 25% compound annual growth over three years in turnover, the same in headcount, or recent R&D relief claims on £1.5m of spending. "Anthropic had seven co-founders with near-equal shares, so none held the required 25% at the outset. In fact, since no more than four people can each own a quarter of anything, no company can ever have more than four deserving founders", notes @grice_maisie. OpenAI also would not have qualified nor would Deepmind. "Founders who had the foresight to pivot into better opportunities may also be undeserving, as the relief requires ‘no major change in the nature of the trade’. Hugging Face, which Nvidia has reportedly agreed to buy for $12.9bn, started life as a chatbot app for teenagers. It shut the app down to build the platform where AI developers share their models". Maybe there is one particular company that Healey had in mind when designing this very narrow relief. But which one is it and why?
22
65
4
184
3,814
Anathemata retweeted
Berkshire Hathaway is now sitting on a staggering $365 Billion in cash, enough to buy 472 companies in the S&P 500 🚨 🤑
76
261
55
1,827
167,507
Anathemata retweeted
Texas ranks ~ third globally in data-center investment, behind the rest of the US and ~ level with China. Kedrosky illustrates that if half of Texas's estimated $66.2B in 2026 data-center spending were disrupted, $9.9–16.6B of Nvidia sales could be delayed, equal to 4.6–7.7% of its FY2026 revenue. edwardconard.com/macro-round…
1
1
254
Anathemata retweeted
🇺🇸 Flow More than $1.7 trillion has flowed into US-listed ETFs so far this year, a record for this point in the year. If the pace holds, inflows could reach $2.3 trillion by year-end, according to model-based projections 👉 isabelnet.com/blog/ @GoldmanSachs #ETF
4
7
1
20
3,842
What if Britain’s grand infrastructure plans are holding back the growth they promise? Perhaps lasting prosperity starts with something less dramatic: properly maintaining the existing systems? on.soundcloud.com/eriJNHgQQ3…
3
8
1
12
2,074
Every southern region just hit a record for how long homes take to sell. London 80, South East 77, East of England 74.
3
1
8
510
Young Britons have almost completely abandoned conservative parties, but it’s not a global trend, per FT In most other countries, including Australia, Canada, France, Germany and Italy, young people back conservative parties at roughly 25% to 40%.
45
26
3
388
140,915