the s-1 that’s now public is also stunningly devoid of numbers. multiple inputs and areas that have literal empty-spaces gaps for numbers

okay felicia

I’d never really looked into the name before but I just now realized what the Korolev was likely named for

yes, chainguard broke embargo, and within a few hours a reverse engineered PoC was kicking. I saw at least two implementations yesterday when assssing impending work

and I haven’t slept yet (although now aaaalmost done…)

there’s also the engineering-side faffery: now you’d need mixed memory controllers, mixed bus speeds, synchronisation issues, etc

simpler to just shove saltman et al in jail and recover world markets

what in the fuck

and I presume the answers default to yes

Betteridge’s reigns supreme

a nsfw to start wednesday, what if the billionaires all suddenly exploded

this is the same company that has attempted other ways of addicting their employees, for “productivity”

ijlth: quantum

I follow a couple (like, 2~3) of fiber arts people and that was about the impression I got too, it’s blown up with that much spill

furthering my previous guessing that there’s likely only appetite for 2, maybe 3, of these ridiculous IPOs, a number of analyst notes have started kicking around and they’re getting progressively more realistic in their outlook. saw this one this morning

Red Flag #3: The IPO is Exit Liquidity, Not Growth Capital

To justify a $2 trillion valuation, our model shows that Anthropic would have to:
 • immediately improve NOPAT margin to 10% and sustain it for the next 10 years (compared to -175% reported operating margin in 2025),
 • grow revenue by 1000% in 2026 and 100% in 2027, and
 • grow revenue 56% compounded annually from 2028-2035.

sounds ambitious. sure hope they don’t have any material risks and systematic exposures that make that harder!

In this scenario, Anthropic’s
1. revenue would reach $3.6 trillion (compared to $4.6 billion in 2025), which is 1.4x the combined TTM
revenue of Microsoft (MSFT), Apple (AAPL), Alphabet (GOOGL), Amazon (AMZN), NVIDIA (NVDA), Meta (META), and Oracle (ORCL).
2. NOPAT would reach $358 billion in 2035, compared to the company’s -$8 billion reported operating loss in 2025. For reference, $358 billion is 2.0x NVIDIA’s $184 billion and 2.6x Apple’s $136 billion in TTM NOPAT.

good thing the cost of inference is going down! sure, qwen3.8-27b is already stale and everyone needs to throw another couple hundred million at another training update, but the cost of inference! /s

personally I feel like all this shit is a continuation/outgrowth from the crypto “whitepaper” phenomenon: ~zero substantive engagement with the actual discipline in question, and “work product” (quotes intentional) equal to some claims with a “proof” thrown out as its defense

(it also continues to be part of a cultural and knowledge DDoS, which others have written better on)