That's honestly pretty rough. Starting a race in a region where it's night time is pointless when you see almost nothing at night. Holding X to summon a vehicle does not seem to work. Can we have less completely broken vibe coded stuff please?
> “Humans of course can drive with just eyes, so there’s that proof of existence,” he said. “If the goal were to just approximately match human performance or to build an assist product, that’s a very reasonable way to go.”
It feels like there's a lot of assumptions baked in here. For example, do humans get in accidents because eyes are insufficient driving sensors? Or because human drivers get tired, distracted, make mistakes, etc.
Also, you can put 10 cameras on a car looking in all directions at the same time, which feels materially different from a human that only has 2 eyes which can only look in a single direction at a time.
It's possible that massively superhuman driving performance can't be achieved with just cameras, but I'm not seeing a strong case for that being made here.
Perf is highly nuanced but we've been doing realistic benchmarking for the ecosystem for almost a year now: https://benchmarks.vlt.sh/
We've got more improvements coming soon but we're a roughly ~10-40% faster registry then npm out-of-the-box. Compared to incumbent enterprise registries we're like ~3x.
Elon may have been technically correct about 'if humans can do it with vision only then theres no reason why vision-only cannot work' but turned out to be the wrong strategy for learning how to get there. lidar gives you a constant stream of data to check your vision yourself against and interatively learn. Once trained you turn it off. whereas teslas strategy ended up having to rely on human drivers for error feedback which made the learning loop way longer.
I am struggling to see the 92% figure as broad operating growth. About $1.9B of the $3.7B increase came from AI infrastructure, but that segment still lost $1.26B from operations and required $15.8B of capex during the quarter.
Consumer Starlink contributed only $764M of the increase, while most important ARPU dropped from $85 to $66.
Unions primarily benefit the weakest workers, and in places like SV, usually the top workers already are getting fat plates and others are pushing hard to get there.
Think of it like a football club holding a vote to flatten player pay. The benchwarmers will be falling over themselves to unionize, and the top players will be looking where to move too.
This is why union efforts fail, which many find perplexing. The most valuable players get better deals negotiating without a union.
It is hard to know for sure, there could be some advantages to more pathogenic exposure in terms of group immunity but it is challenging to study and the particular types of pathogens that get passed around would matter a lot too.
I see nothing wrong with this, neither in principle nor in practice. Also, this was merged over 6 months ago without incident. What is the purpose of this HN submission?
Er...the 50MT device was a cold war dick measuring stunt, not the demonstration of a practical strategic weapon. But, yeah, we (the US) don't even build 1MT devices any more, because 1) we have to fit them into submarines and bombers where size and weight (and economic) requirements limit yield, and 2) we figured out pretty quick that due to diminishing returns on destructive footprint, bigger bombs were pretty much a waste of effort (and produced other side effects like more fallout). Among other gotchas, you're constrained by the cube-root law; ten times more yield is not anywhere close to 10 times more destruction on the ground.
So, since we saw first-hand what devices of ballpark 20KT yield did to an urban environment, numerically most of the operational US nuclear weapons are around 100KT (margin of error and all that)[1]. So, yeah, better targeting, smaller warheads, yada yada. Still one-shot/one-kill for an entire city and the surrounding area. I mean, McNamara could handwave with statements like "optimistically, only millions dead" (paraphrased), but I don't think that kind of thinking has aged well.
That said, things are not worse because the Big Three have Big Bombs (or don't). It's because the strategic landscape has changed. One of the big 3 is lead by a guy who thinks nuclear weapons are just like big hand grenades or something and that negotiation is a zero-sum game. Another bit three leader is maybe just holding on to power by his teeth (or not...who knows) and came of age in the 'nuclear war is winnable' era. And there are more players at the table. Israel has nukes. North Korea. France and the UK. And India and Pakistan. South Africa had them before, and could again. And Iran will get them at some point, sooner or later. And we're apparently going to give the Saudis all the required tooling and expertise required, though there's apparently a firm pinky-swear no weapons will come of it. So we aren't looking at a 'simple' US/USSR cold war, with some Chinese spice in the mix. We have regional power, some of them not known for long-term stability, some with extremely volatile regional situations, with the ability to knock over the apple cart.
So what is the escalation/deescalation path if, say, KYI has a real bender one night and decides that NK nukes Seoul? Or Tokyo? Or what if they just pull a June 1950 and march all the way to Pusan conventionally? Do we nuke Pyongyang? All their nuclear facilities (the ones we know about)? Or glass over the whole place? And what then does China do, us having set off a nuke in their literal back yard on top of their Little Buddy? Under such circumstances, I do not expect the current administration to take a measured step back and say "we need to think this through...".
Uncertainty. That's why a lot of us (who lived in the Cold War) think things look a little shakier than usual.
[1] To head off the inevitable "well ackshually...": Yes, the big players still have big bombs. I haven't been involved in this area in a while so things have probably changed a bit, but we still have some of the big boys in the 300KT+ range sitting on Minutemans for old times sake. I assume the Russians still have a bunch of their 1/2 to 1MT devices in silos. And the Chinese were pretty proud of their absurdly large 4MT+ devices, supposedly still in service.
Atom kinda does. Atom is a paged format, so a publisher can publish their entire history in their Atom feed and only the most recent page gets redownloaded to see new entries as they are added.
Well that's a bad example because a UI isn't a model.
But if you hold that it's true that returns compound and that they do so for intelligence/performance for AI models, then having the lead matters quite a bit and eventually you'll just get into an iPhone vs Android Dollar General situation where sure they do the same thing and one is even substantially cheaper but all the profits go to the iPhone.
If you hold that the lead doesn't matter, then the buildout of data centers and compute capacity which is also what the Chinese are doing will just go to make money for American companies serving whatever open-source models come into being for the rest of however long the world itself lasts. Then China can take the lead and spend a trillion dollars, we'll copy their models (which according to you is a great strategy), and release them open source.
> Going online without an adblocker just feels like playing with fire, especially nowadays.
It feels more like having casual unprotected sex. Easy to do in the moment, easy to ignore the risk, painful if "it never happens to me" doesn't come up cats eyes this time.
If we called the first programs that self-propagated worms and viruses as per the real world analogy, then I think it's time we rebrand adblockers as "condoms".
As for use cases, obviously we can't fully rely on non-deterministic capability for sensitive things but a small model which can do a good job acts as a first defense and then a human can review later.
