theptip

41 ms·

781.
by theptip 2y ago
“Commoditize your complements” is a related point. Content is the input to the Meta ad engine. You can generate it on-platform, or you can bring it in from elsewhere if it is commoditized. (And commoditizing your complements often looks lik
782.
by theptip 2y ago
The point is that none of these guys are risking 100% of anything. Meta stock is liquid. You can sell it or hedge. You might slip 10% but it’s vanishingly unlikely to lose 100%. The well-known tax dodge is to avoid realizing gains by borrow
783.
How Meta enforces purpose limitation via Privacy Aware Infrastructure at scale(engineering.fb.com)
3 points by theptip 2y ago|0 comments
784.
by theptip 2y ago
What do you make of this section that claims recent research has identified an upper bound on the genetic component? > …prediction accuracy depends on sample size, could the findings drastically change with more samples in the future? In
785.
by theptip 2y ago
Under GDPR I believe if the data access is from an employee of the company (eg Uber) then there aren’t location checks. (Been a while so I could be mistaken here.) But if you are subcontracting to an agency you need to list them as Subproce
786.
by theptip 2y ago
As I’ve continued to read I realize the shuttle didn’t have ablative tiles, rather reusable heat dissipation tiles. Better aerodynamics, so maybe our glider should use this approach; the Si and C required are present in asteroids, but manuf
787.
by theptip 2y ago
This is an interesting idea. I initially assumed you’d just structure your payload (eg titanium) to ablate during reentry. But maybe you can do something better with carbonaceous asteroids, or simply with regolith, to produce a cheap ablati
788.
by theptip 2y ago
Scott is debunking ideas to dump deltaV before hitting the atmosphere, to avoid having to go through the rigors of reentry. This isn’t relevant to the discussion; he’s not disproving the possibility of landing from orbit in a glider. The ex
789.
by theptip 2y ago
A medium-sized asteroid identified had something like $3T of titanium on it. Of course, that's gross value and you can’t clear your supply at that price. But even if it costs hundreds of billions to extract you should be able to net a
790.
by theptip 2y ago
Why would you return the materials in Dragon? You can build a simple glider from space-side materials. The return trip is way easier than getting out of the gravity well. You don’t even need crew.
791.
by theptip 2y ago
Tether / skyhook seems a more plausible option.
792.
by theptip 2y ago
I don’t think this is categorically true. A couple obvious options are constructing titanium gliders to land materials on earth, and de-orbiting incoming metal loads with a space tether to provide inertia that can then be used to launch pla
793.
by theptip 2y ago
Isn’t the idea that it’s running the simulation on the GPU; drawing to screen is presumably trivial, the physics is being run per-pixel so there is not much work required to actually draw the scene.
794.
by theptip 2y ago
I tend to agree here. The problem is that many people seem to hold a set of beliefs where sentience == intelligence, and so !sentient implies !intelligent, and conclude that “a stochastic parrot could never outwit a human in the real world”
795.
by theptip 2y ago
Why buy servers if it’s cheaper to hack/steal them?
796.
by theptip 2y ago
I think the salient point is that it was unexpected to the researchers. When people propose regulation for AI systems out of concern for unexpected self-improvement, this is the sort of scenario they are worried about. Security in general i
797.
by theptip 2y ago
The correct policy here is not mandating a “button” or even “the same number of steps to unsubscribe as to subscribe”, rather the government should just mandate adherence to a simple protocol. Then we can just unsubscribe in our mail client
798.
by theptip 2y ago
> we arrive at the interesting number of 8500 GW or 8.5 TW. This power must be sustained by a distributed radiofrequency system, or by an alternative acceleration mechanism. A power source at the level of 10 TW would thus appear to be re
799.
by theptip 2y ago
> One downside of KANs is that they take longer per parameter to train—in part because they can’t take advantage of GPUs. This seems like a big gap. Anyone know if this is a fundamental architecture mismatch, or just no one has written t
800.
by theptip 2y ago
Definitely something to keep an eye on. However, the article does take an a priori negative stance here; > Will Grandma spend her final neglected days chatting with her grandson’s digital double, while her real grandson is mentored by an
801.
Geothermal energy is poised for a big breakout (2020)(vox.com)
2 points by theptip 2y ago|1 comments
802.
by theptip 2y ago
Why would you spin out a competitive moat?
803.
by theptip 2y ago
Strongly agree. I feel this is the recurring problem of armchair philosophers mistakenly thinking they have something to contribute to scientific or political questions. Even among the folks that debate this rigorously, another related prob
804.
by theptip 2y ago
In some dimensions, the current crop can achieve this. See the Google medical AI that gets scored better for bedside manner than MDs. It’s not what we would have predicted pre-GPT, but I think it’s plausible that LLMs will be superhuman in
805.
by theptip 2y ago
Have you actually used a frontier LLM for this usecase? It’s quite different from a search engine. In the case where someone asked your exact precise question including follow-ups, you get the same (or better) results from search. In most c
806.
Oscar, an open-source contributor agent architecture(go.googlesource.com)
175 points by theptip 2y ago|23 comments
807.
by theptip 2y ago
This is one of those modeling exercises where you need a scalar not a Boolean to correctly understand the effect. “Growth over sustainability” is a problem in proportion to how scarce funding is for your company size. (If you have a SoftBan
808.
by theptip 2y ago
Marketing of some sort. Either “come to Google and you’ll have access to H100s and freedom to publish and get to work with other people who publish good papers”, which appeals to the best researchers, or for smaller companies, benchmark pus
809.
by theptip 2y ago
The claim in the OP is that they might have been MiTM’ing arbitrary users, I believe the previously reported claims were that they only MiTM’d paid research participants. (Please share some links if you have evidence to the contrary, I’d lo
810.
by theptip 2y ago
“Cruel and unusual punishment”

More ›