- ...do you not see the parallel in the words you just wrote? That AI's value will continue increasing even as the buzzwords fall out of the public consciousness, just as crypto continued to gain value even as the buzzwords fell out of the public consciousness?
(Paraphrase != endorsement.)
- > This isn't crypto.
Bitcoin is at $92k.
- It's hard for me to respect the intrinsic superiority of a format whose main value-add is exclusivity, rather than fair market competition based on merits.
If theatres pivoted to competing first on format rather than exclusive access to recent releases, and managed to do well in that regime, I'm sure Netflix and other new media would be more than happy to indulge. Seems unlikely, though, doesn't it? The demand exists but I would be surprised if it was a quarter the size.
- Falcon Heavy is a huge outlier, and has never actually demonstrated the capability to lift close to its nameplate capacity to LEO. Falcon 9 is already volume constrained to LEO outside of Starlink or Dragon launches, and Starlink is packed incredibly densely to get to that point. When I ran the numbers some time back, New Glenn was similar to Falcon 9.
Increasing thrust by 15% doesn't just increase payload by 15%. I don't know a simpler way to estimate this than to run a simulation, and I don't have one with numbers I can toggle.
- Tesla got rid of radar because of sensor fusion, and particularly for reasons that wouldn't apply to high resolution radar. Sensor fusion with a high resolution source like LiDAR isn't particularly tricky.
- Lori Garver, former NASA Deputy Administrator, has interesting additional context on the matter.
- Au contraire, the space program stalled because pouring national wealth into gigantic space projects was _too_ sustainable. The idea that NASA has had a lack of funding is a myth. The problem has long been them spending it ineffectively.
- This is as would be expected if it were real. Advantage isn't a black and white thing, because the comparison starts against 'any task done the best we know how to do using the most resources we happen to be willing to throw at it, even if we don't have a means to check that the output was correct', and ends at 'useful output you can formally verify where you have a strong reason to believe no classical algorithm would be effective.'
- > Let’s start with an example: (2#x)#1,x#0 is code from the official q phrasebook for constructing an x-by-x identity matrix.
Is this... just to be clever? Why not
aka. the identity matrix is defined as having ones on the diagonal? Bonus points AI will understand the code better.(!x)=/:!x - Correction: There was flap burn through, just not all the way to the back.
https://bsky.app/profile/dutchspace.bsky.social/post/3m37ofb...
The damage I'd noticed before today was all correlated with where tiles were removed, but the top half of this flap had all its tiles at flight start and still ended up a mess.
- Y'all wouldn't believe this but 10 years ago AGI was a hundred years away.
- > Apollo for comparison took about 8 years from announcement to moon landing. Space shuttle took about 9 years from approval to first flight. Starship has been in development now about 9 years.
This comparison is very unfair. Nine years ago the big rocket was a dream, not even Starship at the time.
> During his presentation, Musk joked that his strategy for raising money might be to “steal underpants,” do a Kickstarter campaign … and profit.
Contra Saturn V, which had strong funding out the gate.
- The hard part of reuse is getting the thing back in flying shape. The booster part is actually just demonstrated, and this flight got us a fully-intact upper stage back, sans a few tiles lost to the wind and the landing being simulated.
I can see room for skepticism on their rapid reuse plans, but skepticism on practicable reuse alone just seems discordant with the demonstrated success.
- Very clean flight, almost all the way through, despite the intentional missing tiles and new flight pattern. No flap burn through, no issues with simulated Starlink deploy, I don't think they even lost any engines.
They're clearly almost ready to scale this thing, if the next block version doesn't add a ton of problems back on. I'm not sure they're quite at the point of rapid reuse looking feasible, since tiles did come loose near the end of flight; not a problem for stage return, but definitely bad enough to warrant a meaningful correction before a (counterfactual) reflight.
Overall they've clearly proven the recipe works.
- This is quite loosely stated. It's true Boston Dynamics is an old company and that they've had some very cool demos. It's not at all true that they've been showing qualitatively similar things for 20 years.
The oldest video on their YouTube channel is 16 years old, and is of a quadrupedal robot not falling over while inching along tricky surfaces.
- There are a few exacerbating factors.
The first is that IIUC, CFCs release chlorine atoms which catalyze ozone, whereas aluminium oxide catalyzes the creation of chlorine atoms from chlorine reservoirs, which then go on to catalyze ozone. I loosely believe at this point after some sketchy research and maths that this makes it around two orders of magnitude more potent.
The second is that these particles are produced directly in the upper atmosphere. I couldn't give you a number for how much that changes things, but I assume it's nontrivial.
The final point I've noticed is that mass to orbit has been increasing at a rapid exponential rate recently, and it would not surprise me at all to soon see an extra order of magnitude on it.
Worst-case, that could change your 5,000 year figure to just a couple. I don't think it's that bad, I'm not overly concerned about this issue, but given ozone depletion is a legitimate existential threat and the numbers don't immediately make it seem impossible, I think it's worth paying attention to.
- This is correct from the perspective of direct health hazards, but there are still plausible risks. We know from history you don't need a lot of mass to cause global problems, if the material is catalytic.
- The intro to the article explicitly states that it will put aside any fanciful ideas that the technology might be better in the future; that "rather than try to forecast the future as it might turn out, I’d prefer to describe reality as it already exists".
The article acknowledges that AI progress to date has worked. It snidely disagrees without argument that AI the field could work from here on out.
- I was thinking more wrt. the top end, which did stall out for 30 years from WWII but has otherwise been on fairly smooth exponentials since the beginning.
https://en.wikipedia.org/wiki/Lindley%27s_paradox#The_lack_o...
Indeed, Bayesian approaches need effort to correct bad priors, and indeed the original hypothesis was bad.
That said. First, in defense of the prior, it is infinitely more likely that the probability is exactly 0.5 than it is some individual uniformly chosen number to each side. There are causal mechanisms that can explain exactly even splits. I agree that it's much safer to use simpler priors that can at least approximate any precise simple prior, and will learn any 'close enough' match, but some privileged probability on 0.5 is not crazy, and can even be nice as a reference to help you check the power of your data.
One really should separate out the update part of Bayes from the prior part of Bayes. The data fits differently under a lot of hypotheses. Like, it's good to check expected log odds against actual log odds, but Bayes updates are almost never going to tell you that a hypothesis is "true", because whether your log loss is good is relative to the baselines you're comparing it against. Someone might come up with a prior on the basis that particular ratios are evolutionarily selected for. Someone might come up with a model that predicts births sequentially using a genomics-over-time model and get a loss far better than any of the independent random variable hypotheses. The important part is the log-odds of hypotheses under observations, not the posterior.