That might be where Intel and AMD failed. With the hindsight watching ARM gobble up a number of sectors, Intel and AMD should have allowed their IP to be customized which may have slowed ARM's advance. Competition is good though. If this continues, AMD and Intel may have to merge. I wonder how RISCV is going to shake things up? Intel and AMD should build RISCV chips and chance ARM. That'll form a nice loop.
Right now RISC-V is ultra slow in all implementations I've seen. Like 30x slower than a top of the line Apple Mx series CPU. Maybe there is a high performing RISC-V chip out there but I haven't yet run into one.
That said, RISC-V is good for embedded applications where raw performance isn't a factor. I think no other markets are yet accessible to RISC-V chips until their performance massively improves.
Google’s AI tool says that 20-25% of the world’s population flies at least three times a year. Not a good source, but at least a surprising statistic if true.
Some hard data says that 12% of US flyers take 66% of flights [1]. Those are all likely very frequent fliers, and is much more than 1%.
At the moment, yes. For a short while, it was very hard to find eggs for less than $5-6 a dozen due to a supply issue. Since then, some egg prices have fallen back to normal. Some egg producers, however, have decided that this new price is just fine for them, so you can still see that price on the shelf.
I have a system where I use a tmux-server config that has a different prefix (^O), status bar color (yellow), and location (top). I use that for my outer tmux session, then start default configs on other host systems. Works pretty well, I don’t have any complaints.
Not in containers by their nature, but you have control over those so you probably won't use amber or bash or perl but just whatever you actually wanted..., but I assure you the underlying hypervisor has perl
Even bash or sh need not be installed in a container so that's a weird objection
The system uses 30MW, but this job used a portion of it that would consume 2.6MW.
There isn't really a figure for how much compute time it takes to train this thing, but 8x H100s have 32PF of AI compute among them. This job had 2,100 (half precision[1]) PF-in-fugaku / 158,956 nodes-in-fugaku * 13,824 nodes-in-job = 182 PF-in-job, implying it can get the job done 5.6x faster, or a little over ten days at the most optimistic.
Electricity costs for these nodes for ten days looks fairly similar to the rental costs of 8xH100s for 60 days according to my research. Lambda labs seems to have very cheap instances for 8xH100, but AWS and its ilk are much higher. However, the comparison is a little weird, as Fugaku is also a few years old now, and the contemporary GPU at the time of its release was the A100 (1/13th of an H100). The next Fujitsu chip may well narrow the power/performance gap between itself and (say) Blackwell or whatever is current at the time.
Undertaking a claim that there is no one you can speak to Facebook about this will help them quickly find someone to talk to you and respond to the credit card company and you.
When standard channels aren't working, complain to someone who can do something about it.
A couple of times in my life, I've had good luck contacting the office of Chairpman of the Board. I had problem with a debt collector of American Express, I was a financial delinquent in those days, and I contacted American Express directly. I asked for the Chairman of the Board's office, and talked to his assistant. I explained the situation about how I was being lied to by the debt collector, the assistant arranged to have the debt recalled to American Express, and I worked w/ Amex directly after that.
Another time, also w/ Amex, my friend's consulting invoice wasn't being paid my Amex in a timely manner. I recommended she contact them using the same approach, and an assistant helped her get paid.
More recently, I had a problem getting my tax return processed with the Internal Revenue Service. I complained to my Congressman, his office put me in touch with a tax advocate at the IRS, and everything was cleared up. This is the second time I needed to do this.
Yeah, disputing the charge might be better than a straight chargeback. The merchant gets to respond to the dispute and provide evidence, and if they clearly deny having issued the charge then it gets reversed and you can move on.
I’ve just come to realize from this article that human challenge trials are really a form of trolley problem [1]. Is it more ethical to “do nothing” and gather data from infections in the wild, where many people are potentially harmed, or do you “throw the switch” and infect a much smaller number deliberately to get high quality data?
The labor market (how to get money if you don't have enough) creates a perverse incentive. I wonder if anyone has reaearched the "volunteers" and their position in both labor and wealth.
I wonder why I never hear this argument applied to government spending in general. In non-authoritarian countries, spending money is the primary means for the government to inflict their will on the public, regardless whether it will have a positive impact on society.
I think it's a question of "treat the symptom" sometimes as well. Some people really cannot make ends meet due to their locality and don't deserve to be stiffed by a migratory economy.
they are being forced by statistics. It's a supply/demand problem. If you get people in a bad economical situation and then give them enough money you will get them to do your thing. Just increase the amount of money until you get enough people signing up.
> Now picture what would have happened if we had been willing to do challenge trials early on for COVID.
Nothing significant would have changed. For example, on July 22, 2020 HHS announced $1.95B in funds for Pfizer for large scale manufacturing and distribution of 100 million doses of their vaccine. On Nov 18, 2020 the phase 3 clinical trials were completed. We didn't wait to know that they worked or not to start ordering the doses. It still took until April or so of 2021 to get that manufacturing and distribution completed because we'd never made nanoparticle vaccines commercially before at all and distribution itself was hard once we had the doses.
An issue with doing a COVID-19 challenge trial that I heard from someone in this space at the time: Nobody actually knew how much virus to administer. We weren't sure of the normal quantity of COVID-19 a person typically inhales before becoming sick.
A major argument in favor of a challenge trial was that, for people who are young and otherwise healthy, COVID-19 isn't particularly deadly. However, we don't know what would have happened if we accidentally gave participants 10x the normal dose of COVID.
>An issue with doing a COVID-19 challenge trial that I heard from someone in this space at the time: Nobody actually knew how much virus to administer. We weren't sure of the normal quantity of COVID-19 a person typically inhales before becoming sick.
I dont think that is accurate. You dont have to know the actual viral quantity transmitted to create a representative transmission event. That is to say, if you know people can catch covid sitting side by side, that can be your challenge.
Even if the scenario isn't perfect, you still know how many people caught it vs placebo.
Yes, but again: If you gave participants an ineffective or placebo vaccine followed by 10+ times the normal dose of COVID, how many of them would die?
If you're okay with potentially killing most of the trial participants, I suppose you could still get useful data, but the ethics become significantly more questionable IMO.
Edit: I just re-read your post. I think you're saying, you wouldn't actually give anyone the COVID virus directly, you'd just find someone who was known to have contracted COVID in the wild and bring them in to purposefully expose trial participants. That is an interesting idea which I've never seen discussed in the context of a challenge trial. I'd be interested to read more about why this is not typically done.
Getting a 10x Lethal dose just doesn't seem like a realistic concern to me, and certainly not a dealbreaker.
We know and knew what a normal exposure was: being in close proximity to one or more contagious people for somewhere between a few minutes and many days.
If you have someone sit next to an infected person for an hour, how do you get to this hypothetical 10x super dose?
Edit: I would also add that medical trials also have well established methods of determining effective and dangerous exposure levels, and these are used in almost every Phase 1 trail. You start with extremely low exposure, then increase it until you see dangerous outcomes. This could easily be done with a covid serum to find the infectious but non-lethal level. for example, if you think X amount of covid virus is infectious, you start at 0.0001 X and then increment up, to find the amount that causes normal infection, and not some lethal mega dose.
> Edit: I would also add that medical trials also have well established methods of determining effective and dangerous exposure levels, and these are used in almost every Phase 1 trail.
Yeah, so the scientist I heard this from didn't say it would be impossible to find the right dose, just that we didn't know the dose yet and we'd have to find it before running a challenge trial.
He thought this erased the time savings of a challenge trial versus the normal process.
> Now picture what would have happened if we had been willing to do challenge trials early on for COVID.
Nothing much most likely. The mRNA vaccines were designed very rapidly after the virus itself had been sequenced. The thing that took time was moving through the different trial phases. Given that there was more than enough spread of the virus in the wild, deliberately exposing people might have shaved off a few weeks at most.
But the real bottleneck afterwards was production and roll-out of the vaccines anyway. So realistically, challenge trials would not have had any meaningful impact.
>The thing that took time was moving through the different trial phases. Given that there was more than enough spread of the virus in the wild, deliberately exposing people might have shaved off a few weeks at most.
I dont think that is accurate. You can run a challenge trial from exposure to outcome in weeks, whereas it takes many months and tens of thousands of people to get enough cases in the wild.
It may still be that production was the critical path, but challenge trials are much faster.
In the case of COVID-19, for persons circulating in public, exposure was pretty much guaranteed during peak periods of the pandemic, most particularly (Northern Hemisphere) Summer 2020 and Winter 2020 (June -- August and November -- February, mostly). Challenge exposure would have been largely redundant, though it might well have been a utilised protocol.
The time constraints were the sequencing of multiple innoculations (most vaccinations require at least two doses spaced at 2+ weeks for preferred effect, with additional subsequent boosters increasing effectiveness), time from final innoculation to infection, observing course of illness, if any, and monitoring for any fall-off in immunity over time. Even on an accelerated basis that's a 90-day period, roughly, and longer-term assessments of 1--2 years are of clinical interest. Though once it became clear that vaccination showed greatly-increased immunity within weeks of treatment, authorisation of the vaccine was virtually assured.
At which point ramping up production and distribution were the challenges.
The initial mRNA vaccines were prototyped within days of the SARS-COV-2 virus being sequenced. It took nearly 18 months to go through the testing, production, and distribution of the vaccine to the point that anyone who wanted a vaccine could obtain one. Little of that lag had much to do with trial phases, though identifying better treatments (several of the vaccine lines proved less effective, notably the Chinese-created vaccine AFAIU) also had value. Parallel development and testing was effectively pursued in this case.
It contains the tension between utilitarianism and the deontological act/omission distinction that the trolley problem is meant to isolate, but clinical trials are philosophically very different in that (1) the risks you cause during a trial are statistical rather than leading reliably to death of any one person, (2) trial participants can and are compensated for this risk, and (above all) (3) the trial participants have consented. In other words, there very little tension/inconsistency in saying that throwing the trolley switch is unethical but running human challenge trials is ethical.
Obviously the covid vaccine was an unprecedented effort, but do remember reading how the high rate of infection in the US made it statistically easier to test. Within a few months you could observe the gap between the control and treated group.
With vaccines for things like cancer, that take decades to develop and at a low rate, you need to wait a very long time even if you have a large group.
I think that most people would say that diverting the trolley to where you, yourself, are tied up would be a moral but supererogatory thing to do. But we still make it impossible to do that because the relatives of the person who sacrifices themselves might sue the institution that allows it.
reply