jsnell

11 ms·

91.
by jsnell 5mo ago
The AI row of the capex table in the S-1 should be a pretty close approximation.
92.
by jsnell 5mo ago
The cost of power is pretty much irrelevant. The TCO of an AI datacenter is dominated by the capex. Over the lifetime of a DC, the capex will be 10x higher than all the opex combined. Power matters, but what matters is power availability no
93.
by jsnell 5mo ago
The HN guidelines specifically ask you not to do what you're doing, and say what to do if you have a genuine concern: > Please don't post insinuations about astroturfing, shilling, brigading, foreign agents, and the like. It de
94.
by jsnell 5mo ago
No? I think you're misunderstanding what is being measured. It is purely a test of capabilities (can it do a thing that takes a human $X hours), not efficiency (how fast will it do it).
95.
by jsnell 5mo ago
We don't know. There is reporting that they have positive EBITDA, but that's not the same as profitable. Basically it would completely ignore launch costs. They also aren't just the rocket and satellite company any more, but
96.
by jsnell 5mo ago
Do you have any part examples of them committing to a specific support timeline on a product and reneging on it? I can't think of one.
97.
by jsnell 5mo ago
It is particularly funny because this is content marketing for a computational proof of work "captcha". Those are pure snakeoil, with economics that are probably at least four orders of magnitude more favorable to the abusers than
98.
by jsnell 5mo ago
You can check https://everyuuid.com/ for collisions.
99.
by jsnell 5mo ago
This doesn't seem to be controlling for the number of turns in any way. Am I missing something? Stronger models needing fewer turns to achieve a task feels like a prime source of efficiency gains for agentic coding, more so than indivi
100.
by jsnell 5mo ago
Because they are >10x more power efficient. If silicon were relatively abundant and power/DC space scarce, you'd get an order of magnitude more bang for the Watt by replacing the H100s with newer GPUs. But nobody is doing that.
101.
by jsnell 5mo ago
All of that is incorrect. H100 rental prices are still as high as when the cards were brand new. The prices vastly exceed the power costs. In a world where power or DC permits are the current bottleneck those H100s would be getting retired
102.
by jsnell 5mo ago
The current bottleneck is silicon. Every chip that is manufactured gets housed and powered. (It makes sense: the cost of compute is dominated by capex, the power costs are irrelevant, so they're ok paying a premium for power). The sp
103.
by jsnell 5mo ago
Right, where is the rest of the code?
104.
by jsnell 5mo ago
The Google ad network's revenue is 10% of their first party ad revenue. It would be even harder to make the numbers work that way.
105.
by jsnell 6mo ago
If you try to track down the actual source for that Tom's Hardware link, it becomes pretty obvious that the claim is not credible. [0] GPUs do not burn out in three years, H100 rentals are priced at the same level as two years ago, and
106.
by jsnell 6mo ago
Try updating your Claude Code client. I believe it is a bad interaction between Opus 4.7 and older system prompts.
107.
by jsnell 6mo ago
I don't see how that number could possibly be realistic. A H100 cost 30k when new, and uses 500W of power. 500W for a year is about 4500kWh, which at $0.10/kWh is $450/year if run at full utilization (unrealistic). TCO of an
108.
by jsnell 6mo ago
> Pick any two cypherpunks at random and you won't find that kind of overlap on non-technical quirks. That could be a valid methodology if you pre-registered the list of quirks before doing the investigation. But in this case the jo
109.
by jsnell 6mo ago
Obviously no guarantees that it's exactly what was done in this case, but he talked about his general process recently at a conference and more in depth in a podcast: https://www.youtube.com/watch?v=1sd26pWhfmg https:
110.
by jsnell 6mo ago
> Do you have a link to that? A rather important piece of context. It was a quote from your own link from the initial post? https://www.freebsd.org/security/advisories/FreeBSD-SA-26:08... > Credits: N
111.
by jsnell 6mo ago
Except it's not 100x revenues, and it's not 17% growth. I don't know where you got those numbers from? The numbers OpenAI gave in the post would mean a 30x multiple pre-money. And the $20B -> $24B run-rate growth since the
112.
by jsnell 6mo ago
What what? Are you surprised it's that low, that high, that they can tell what their revenue is, that they report it on a monthly rather than annual basis, or something totally different? It's going to be pretty hard to get a good
113.
by jsnell 6mo ago
A theory that at least is consistent with the observed correlation seems vastly superior to a midbrow dismissal that doesn't. Your "raising kids is hard" theory would explain why people don't have a third child, but rais
114.
by jsnell 6mo ago
Right, but the business decisions probably aren't the constraint at this point? (But were a year ago.) Once the ability of the supply chain to grow has been saturated, no amount of extra confidence will make it grow faster.
115.
by jsnell 6mo ago
A new fab will need to be filled with advanced equipment like lithography machines. They are the most complex thing humanity has every built. There is one supplier of EUV lithography machines in the world, ASML. They are basically acting as
116.
by jsnell 6mo ago
It wasn't actually that exact amount. It was "about 12 tons", and somebody did the 12000 kg / 29g calculation and used the answer with way too many significant digits. Probably the reporter trying to make the 12 ton numb
117.
by jsnell 7mo ago
On Chrome, right-click brings up the context menu in addition to flagging a mine, which basically makes this unplayable. Also the ship is not explained at all (the graphics, the controls, the systems). I'd recommend at least a one para
118.
by jsnell 7mo ago
I think you've misunderstood something. This is not about rejecting LLM-written articles. It is about rejecting the articles of people who used LLMs for their reviews. So your quip is just nonsensical.
119.
by jsnell 7mo ago
Almost certainly so that they can afford more AI data center capex.
120.
by jsnell 7mo ago
A practical question: what should readers do when they suspect a comment (or story) is AI-generated? Is that an appropriate reason for flagging? Email the mods? Do nothing? I've been pretty wary about flagging AI slop that wasn't

More ›