Vastly more favorable today than it was when Concorde flew.
1) Rich people are WAY richer, and time is even more valuable
2) Businesses have some very important employees and "2 day trip" vs "3-4 day trip" is worth $50-100k
3) Larger population of people able to pay $20-30k for a flight than ever before.
The biggest practical impact is there's probably going to be a private jet version instead of just a commercial one, and there will likely be transpacific demand exceeding transatlantic. Also government/military use.
You are not thinking high enough the food chain. I mean heck you have tenured SV engineers cracking $1mm with RSUs. It’s not rare in finance for folks to be hitting $3-5mm with bonus. So that’s what $19k comp a day. If that individual is making $5mm they are more than likely making a multiple of that for the organization.
Even these days, a lot of retailers operate fleets of private jets even for district or regional managers, because it saves somebody like Walmart a lot of paid hours to fly someone from rural town A to rural town B rather than potentially deal with the hassle of an overnight booking at an airport hub.
I still don't quite get it given I have never worked high enough or in a big enough corp. What kind of mission a person earning $19k a day have to do at the destination to justify the cost? I imagine to earn this much their main responsibility is to lobby / influence someone important (at dinners, golfing and such). Otherwise, if there are no outsiders involved the whole thing could be just done online. If it's about lobbying though - does it have to be done immediately and 3-4 times a week?
Another example that comes to my mind is a highly skilled expert in repairing some important machinery, e.g. ship engines or factory lines.
Again thinking too small. Large business dealings are still done in person, who is going to give another entity $50mm or more without meeting in person, having dinner and getting to know the counter party. It’s not about lobbying so much as time in person counts.
Who said anything about 3-4 times a week? There are plenty of businesses with high earners where time matters and I could see flights like these being profitable. Boards meet quarterly typically and they are often preferred to have in person.
Well yea great but why do they have to be there so quickly that not even a private jet is good enough?
What kind of multi million dollar deal blows up because a dude arrives 18 hours later? And what are they doing when they get there that couldn’t have been done online?
Again, time is money. It’s not about arriving later but the efficiency of time. Many different attributes to consider from sleep, to time optimization or more simply a lot of folks don’t particularly enjoy the long travel and it’s worth the premium. The kind of trips I am thinking about are business related, may only last a day or two with zero buffer before or after.
In business face time still matters. Less than prior to COVID but it still matters. Most boards prefer to meet in person.
If a fund can get into the Anthropic Series A vs. not get into the Anthropic Series A based on a flight to meet with the team, that is worth buying the supersonic plane even if you scrap it after.
Yeah, my first impression when I saw this was: if this is accurate, the situation is not nearly as bad as I thought.
I do wonder why Nvida is included, though. If you include the company that all of the frontier models are pouring money into, of course the net (expenditure - profits) of the collective is going to be closer to zero :-)
If Nvidia is included, does that mean that the money Amazon, Microsoft, and Oracle get for selling compute to the frontier models are included in their revenue?
Because for Amazon in particular, the situation this pages shows is actually much WORSE than I expected. I thought they were making a killing selling compute for model training.
Right, especially given that majority of this investment is into GPUs and data centers that are amortized over a longer period of time. This is actually very hopeful.
Given how the curves look like in terms of ramping of spend, these are very healthy numbers.
The critic I see most frequently on the unprofitability of AI is Ed Zitron. I am sincerely curious if he shorted Facebook's, Amazon's, or Google's stocks. Or if he's in index funds which have tech stocks like those.
For example: I have index funds which have some of these stocks. So I, by process of revealed-preference, don't think it's a bubble, or I think I will keep my money in through the bubble's pop. I don't have that much else to say!
For the record: I would respect the creator of this site equally or more if he/she said, "I'm shorting these stocks and this is why."
Oh really? A 195% cost to revenue ratio isn't bad at all? I'm not a biz expert, but I spent a few minutes looking this up (e.g., what are usual cost-to-revenue ratios for new lines of business), and this sounds like BS to me.
If "cost" is mostly capital investments, absolutely. Normally you'd use operating cost (which for capital equipment would be depreciation and interest), and here they are using the capital cost as full cost.
No one really knows how quickly AI hardware investments will become obsolete and thus how long it should be amortized, but 2-3 years would be extremely conservative, and in fact used H100 (discontinued/2 generations old) prices are higher today than they were when the equipment was new several years ago.
But if it's fully being amortized, then it means they don't buy new Nvidia GPUs anymore for a while. The situation is either "your GPU AND the datacenter infrastructure it's running on is obsolete", or "Nvidia's profits tank because people are staying with current-level infrastructure".
That would be true if everyone weren't supply constrained and buying literally everything they can find.
There are actual risks that this trend doesn't continue, but as long as the trend continues, it is pretty good for revenue. "AI shown to hit a wall/doesn't actually deliver/stops growing so fast", "massive improvement in hw efficiency or tech such that all the old stuff becomes obsolete", "bottleneck on power/regulations/etc such that no one wants anything but the most efficient cutting edge stuff" would be the ways it could end and then all these factors reverse. Right now, power is so constrained that old, inefficient power generation is actively being turned back on or set up at new sites (e.g. old aviation turbines which are very inefficient compared to combined cycle).
I think RISKS (which he edited) was one of the first places I saw computer security discussed in a meaningful way outside a military context (NSA, Puzzle Palace, Rainbow Books, etc.)
Where I live (Puerto Rico) Costco is basically the only decent midrange to upper midrange store. There are a couple higher end butcher shops, but local groceries are "smart and final" or other discount-US grocery equivalent, with prices at mainstream NYC levels, while Costco is basically the same prices as most of the US, and ~70% of the selection of Bay Area Costco. Easy choice (along with Amazon and other online shopping).
If the issue are tent stakes/lag bolts which get buried under surface, clear solution would be metal detectors available to borrow/rent (or brought by each camp). Also probably could do a drone or ground robot with a metal detecting loop on the bottom.
The best solution I know of is to get three-link segments of chain and put one on each screw as it goes into the ground. That not only marks the spot, it also gives you a flexible attachment point which is useful in all sorts of situations. (Two links would be pinned in a stationary fashion.)
Biggest problem is it’s a pain in the ass to chop up all that chain, and nobody sells them in pre-cut lengths.
Closest I've been to losing vision in one eye was creating these 3x chain links for Burning Man.
Naive thought: I could use a large bolt cutter to cut chain links. Started trying to cut a link, felt it was sketchy, went and put on some safety glasses.
Restart cutting (had these bolt cutters with like 1m long arms), apply full force, jaws slip a bit on the chain, jaws bite hard. Chunks of steel fly into my chin and face, metal chunks embedded in chin, cracked safety glasses. Dodged a bullet.
Ended using a small welded up jig so I could stretch the chain and then use angle grinder to cut the chain links. Still sketchy, but no flying metal chunks.
He was pretty shockingly an entrepreneur and inventor in all the best ways,’in a field dominated by very cautious scientists (who are great too, but who likely never would have gotten the genome sequenced within 10-20 years of when he did it). It was basically the Apollo Project in a field which was more like 1980s NASA in culture.
> in a field dominated by very cautious scientists (who are great too, but who likely never would have gotten the genome sequenced within 10-20 years of when he did it).
I did a bio undergrad and one of my professors was involved. She was adamant that the Human Genome Project finished ahead of Celera and that the HGP published reference data that Venter and team fundamentally relied upon to even have their shotgun approach work.
i worked for ham smith and my understanding through him is that both sides relied on data that the other produced.
here are technical details, both were more or less independent, the celera sequence did include data from the other side as useful reference points but the assembly would have happened without it.
https://pmc.ncbi.nlm.nih.gov/articles/PMC123615/
iiuc it was hamilton smith who insisted that shotgun sequencing would work. the nih side insisted on primer walking until celera started assembling the genome so rapidly that the nih had to get in on shotgun too
I belive you are mixing assembling the genome by combining sequences of individual, overlapping inserts of cosmids, fosmids, PACs and BACs (bacterial vectors with human DNA inserts of 40-150kbp) to whole genome shotgun.
The inserts of the above bacterial vectors were sequenced using shotgun, but the gaps in the sequence were closed with custom primers.
My theory is a lot of the anti-AI sentiment is specifically US geopolitical adversaries (pick one or more: China, Russia, Iran, ...) who want a bad outcome for the US (AI as potential AGI; AI as one of the few successful economic sectors of the US; general desire to cause societal disruption or collapse and AI as convenient target). Probably >95% of the really bad stuff (the micron fab disruption, attacks on AI datacenters, ...) is probably root-cause that, possibly executed by useful idiots, people paid by organizations, etc. 5% is normal NIMBY stuff. Approximately measure 0 is Zizian death cultists.
I don't any of these will be dissuaded by cute family photos. Fortunately the frontier model companies and major infrastructure providers are able to pay for top-tier corporate security (although tech people generally have been unwilling to do this at home for lifestyle reasons), but I'd be afraid for people elsewhere in the supply chain.
(And destructive attack is all on top of the normal corporate espionage, infiltration, subversion, etc.)
If a good outcome for the US is OpenAI technology being used by the US military to kill Middle Eastern children, I want a bad outcome for the US too. (Proudly born and raised in California)
It's depressing now, but also was genuinely amazing how great EFF was early on. I think a lot of that had to do with the board, membership, and staff (such as yourself) intentionally trying to keep things balanced and focused. Thank you for all the great stuff you and the rest of the org did back then.
It will be interesting to compare PQ rollout to HTTPS rollout historically (either the "SSL becomes widespread in 2015" thing, or the deprecation SSL 3.0). Cloudflare is in an easy position to do stuff like this because it can decouple end user/browser upgrade cycles from backend upgrade cycles.
Some browsers and some end user devices get upgraded quickly, so making it easy to make it optionally-PQ on any site, and then as that rollout extends, some specialty sites can make it mandatory, and then browser/device UX can do soft warnings to users (or other activity like downranking), and then at some point something like STS Strict can be exposed, and then largely become a default (and maybe just remove the non-PQ algorithms entirely from many sites).
I definitely was on team "the risks of a rushed upgrade might outweigh the risks of actual quantum breaks" until pretty recently -- rushing to upgrade has lots of problems always and is a great way to introduce new bugs, but based on the latest information, the balance seems to have shifted to doing an upgrade quickly.
Updating websites is going to be so much easier than dealing with other systems (bitcoin probably the worst; data at rest storage systems; hardware).
If any kind of proof about serious quantum computers comes to light, browsers can force most websites' hand by marking non-PQ ciphers as insecure.
Maybe it'll require TLS 1.4/QUIC 2, with no changes but the cipher specifications, but it can happen in two or three years. Certificates themselves don't last longer than a year anyway. Corporations running ancient software that doesn't support PQ TLS will have the same configuration options to ignore the security warnings already present for TLS 1.0/plain HTTP connections.
The biggest problem I can imagine is devices talking to the internet no longer receiving firmware updates. If the web host switches protocols, the old clients will start dying off en masses.
Leaf certificates don't last long, but root CAs do. An attacker can just mint new certs from a broken root key.
Hopefully many devices can be upgraded to PQ security with a firmware update. Worse than not receiving updates, is receiving malicious firmware updates, which you can't really prevent without upgrading to something safe first.
> An attacker can just mint new certs from a broken root key.
In Chrome at the very least, the certificate not being in the certificate transparency logs should throw errors and report issues to the mothership, and that should detect abuse almost instantly.
You'd still be DoSing an entire certificate authority because a factored CA private key means the entire key is instantly useless, but it wouldn't allow attacks to last long.
When you connect, you specify supported ciphers. If the server doesn't support them, there's standard "insufficient security" (71) error that was there since at least TLS 1.0, maybe earlier.
They are slower, larger, and less tested. Specifically the hope was to develop hybrids that could also provably be more pre-quantum secure then what they are replacing. History dose not favour rushing cryptography.
They are large, but they're not that slow actually. We've been testing them for almost a decade now. I agree that rushing is bad. That's why we need to start moving now, so that we're not rushing even closer to the deadline.
Waiting now means rushing even more close to the deadline! We added stats on origin support for post-quantum encryption. Not as much support as browsers of course, but better than I expected. Still a long road (and authentication!). https://radar.cloudflare.com/post-quantum
> Updating websites is going to be so much easier than dealing with other systems (bitcoin probably the worst; data at rest storage systems; hardware).
Does it? That one is different because IPv4 with CGNAT largely "just works" except for P2P type stuff. As a result there's a strong incentive for anyone who has a working setup to just not care.
I can use myself as an example here. IPv6 is supported by all my hardware, all the software I use, and my ISP provides it. Yet my LAN intentionally remains IPv4 only with NAT. Why? Because adding IPv6 to my LAN would require nonzero effort on my part and has (at least for now) quite literally zero upside for me. If I ever need something it offers I will switch to it but that hasn't happened yet.
PQC is entirely different in that the existence of a CRQC immediately breaks the security guarantee.
I insure a bunch of big datacenters (crypto mining, AI); there are really two main drivers of cost of insurance per $ of equipment:
1) Internal risks and controls within the datacenter (the company involved and their operating history, fires, flood, etc,) -- for a sufficiently "good" datacenter, you can assume it gets maxed out in quality, or at least to the point where it's no longer efficient to spend more. Most of these risks also cause service disruptions, so if you're building for high availability anyway, the rest of this stuff is usually handled as part of that. Essentially, if you're too cheap to build a good enough datacenter to max this out, you're not getting insurance anyway in most cases, so it's not a variable factor so bunch as binary or maybe a few broad risk bands (ISO tier for datacenters).
2) External risks. This is mostly natural catastrophe ("nat cat" or "cat risk"); usually there's one dominant driver of that ("severe convective storms" in Texas; floods and hurricanes in places like Florida; earthquakes in California). In some places it's multiple risks (Japan has both earthquake/tsunami and typhoon). This drives the majority of insurance premium.
War risk, geopolitical, political risk, terrorism, SRCC ("strikes, riots, and civil commotion") are in a third category -- often essentially not a factor (e.g. for a $200mm facility in rural Texas), but often handled through special programs at a national level or specialty insurance. A lot of normal policies exclude or let the client buy-back that part of the risk.
As my personal interests in war zones, drones, etc. and professional interests in crypto, AI, and datacenters seem to have converged, looking forward to seeing "quality of air defense artillery/integrated air defense system" as well as "comprehensive quick reaction force capable of dealing with national-level threats" as elements of insurance underwriting for $50B AI datacenters/"AI factories" in the future. I assume in most cases this kind of stuff will be handled by national, military, defense, or civil defense parts of the government, but could easily be contracted as well. I don't think Oracle Cloud is likely to stand up their own private army though.
1) Rich people are WAY richer, and time is even more valuable 2) Businesses have some very important employees and "2 day trip" vs "3-4 day trip" is worth $50-100k 3) Larger population of people able to pay $20-30k for a flight than ever before.
The biggest practical impact is there's probably going to be a private jet version instead of just a commercial one, and there will likely be transpacific demand exceeding transatlantic. Also government/military use.