> Once the codebase has become fully agentic, i.e., only agents fundamentally understand it
What exactly do we mean this? Because it is obviously common for human coders to tackle learning how an unfamiliar and complex codebase works so that they can modify it (new hires do it all the time). I can think this means one of two things:
* The code and architecture being produced by agents takes approaches that are abnormally complex or inscrutable to human reviewers. Is that what folks working with cutting edge agents are seeing? In which case, such code obviously isn’t beeping reviewed; it can’t be.
* the code and architecture being produced by agents can still be understood by human reviewers, but it isn’t actually being reviewed by anyone — since reviewing pull requests isn’t always fun or easy, and injecting in-depth human review slows everything down a lot — and so no one understands how the code works. (I keep thinking about the AI maximalist who recently said he woke up to 75 pull requests from his agent, like that was a good thing)
And maybe it’s a combination of the two: agent-generated pull requests are incrementally harder to grok, which makes reviewing more painful and take longer, which means more of them go without in-depth reviews.
But if your claim is true, the bottom line is that it means no one is fully reviewing code produced by agents.
This is classic Quanta Magazine sensationalism. Here's what the study actually said:
As cells in epithelial tissue get crowded, their membranes start to allow more sodium ions to enter, which makes the cell more electrically positive (depolarization). The cells try to counter this, but cells with insufficient stored energy (ATP) will struggle to do so, and will lose water through their membranes, causing them to shrink, which causes them to signal their neighbors to extrude them.
So there's no "group decisions" being made, no "coordination" between cells using "bioelectricity". Yes, all cells rely on electrical potentials across their membranes for normal functioning, potentials that they have to maintain. That's all the involvement of electricity here.
And the only "decision-making" happening here is within a single cell, but of course cells don't "make decisions', cells are little machines, and part of the mechanism for epithelial cells -- a mechanism that works in part using chemistry and electricity -- includes the cell signaling that it needs to be extruded in certain circumstances, like shrinkage.
> I don’t believe genetics ever claimed to provide a theory of why eyes grow where eyes grow.
That’s the whole point of developmental biology, to show how features of the human body form and develop based on gene expression, the timing of which during embryonic and fetal development itself is dictated by your genes.
If not your genes, what else would determine why you have eyes in about the same place in your head as every other human?
> The cells in your eyes have exactly the same DNA as the cells in your big toe, so developmental morphology cannot be explained with DNA alone.
Sure it can, because while every cell has essentially the same DNA, the expression of genes differs between cells, which is what causes cells to differentiate. And this differentiation also controls development; look up the Hox genes as an example.
> That "why" is almost missing from the public conversation. People jump straight to hardware and hand-wave the business case, as if the economics are self-evident. They aren't.
But then he never answers that fundamental question, and jumps straight to the hardware and power and cost? What problems are orbital data centers trying to solve? What optimizations are they intended to deliver? Are these optimizations beneficial to everyone who uses a data centers, or just operators or users of orbiting satellite constellations?
> But the knock-on effects are why this keeps pulling at people. If you can industrialize power and operations in orbit at meaningful scale, you're not just running GPUs. You're building a new kind of infrastructure that makes it easier for humans to keep spreading out. Compute is just one of the first excuses to pay for the scaffolding.
This seems to be the closest we get to a “Why”, but it doesn’t make much sense. A constellation of 40,000 satellites with GPUs “infrastructure that makes it easier for humans to keep spreading out”? How?
> The target I care about is simple: can you make space-based, commodity compute cost-competitive with the cheapest terrestrial alternative? That's the whole claim. … Can you deliver useful watts and reject the waste heat at a price that beats a boring Crusoe-style tilt-wall datacenter tied into a 200–500 MW substation?
Isn’t the answer clearly “No”? The default settings of his model — which I assume he considers optimal — tell us that power for orbital data enters will cost 3.5X terrestrial ones. And that only SpaceX has the vertical integration to do even attempt to do this. So again, where is the competitive advantage?
Also, I don’t understand why he’s including satellite construction and launch costs for a 40,000 satellite constellations in his analysis, if he’s assuming SpaceX as he claims. Wouldn’t SpaceX simply implement these compute capabilities in the next gen of Starlink, so which would reduce costs significantly.
> It might not be rational. But it might be physically possible.
But isn’t that precisely what everyone has been saying? I don’t think the question has been whether orbital data centers are possible, it’s been whether they are rational. And that centers foremost h the unanswered question, Why is this a good idea?
> Someone with a subscription logs into the site, then archives it.
That’s not the case. I don’t have a NYT subscription, I just Googled for an old obscure article from 1989 on pork bellies I thought would be unlikely for archive.today to have cached, and sure enough when I asked to retrieve that article, it didn’t have it and began the caching process. A few minutes later, it came up with the webpage, which if you visit on archive.is, you can see it was first cached just a few minutes ago.
My assumption has been that the NYT is letting them around the paywall, much like the unrelated Wayback Machine. How else could this be working? Only way I could think it could work is that either they have access to a NYT account and are caching using that — something I suspect the NYT would notice and shutdown — or there is a documented hole in the paywall they are exploiting (but not the Wayback Machine, since the caching process shows they are pulling direct from the NYT).
> So can someone who owns a modern car please help me understand why you would buy a car that has the mere capability to be remotely shut off?
That’s not what is going on here. These cars are not being intentionally shut down remotely. Instead, a software update for some computerized components of the car was pushed down to the cars and installed with the owners permissions, but that update apparently has severe bugs that should have been caught by QA.
I don't see how that's an apt analogy. Geocentrism put the Earth at the center of the universe, around which the rest of the universe rotated. But saying life on Earth originated on Earth does not in any way put the Earth at the center of anything. Nor does it in any way mean that Earth is unique.
The bottom line is that -- because we don't know how abiogenesis occurred, whether here or somewhere else -- we have no way to judge how common it is. It could be that, given enough time, life spontaneously forms on any planet or moon that offers a certain set of conditions, and Earth just happens to be one of those planets, meaning it is still not "the center" of anything.
In fact, in the extreme case, panspermia is much more geocentric, saying that life formed in just one very special place -- maybe not the Earth, but somewhere, the "center of life in the universe" -- and then spread by diffusion to all the other locations in which life existed. But that seems like an unlikely and unnecessary model; if life can spontaneously begin somewhere, why should we assume it can't begin in many places, and if that's true, why not also on Earth?
What exactly do we mean this? Because it is obviously common for human coders to tackle learning how an unfamiliar and complex codebase works so that they can modify it (new hires do it all the time). I can think this means one of two things:
* The code and architecture being produced by agents takes approaches that are abnormally complex or inscrutable to human reviewers. Is that what folks working with cutting edge agents are seeing? In which case, such code obviously isn’t beeping reviewed; it can’t be.
* the code and architecture being produced by agents can still be understood by human reviewers, but it isn’t actually being reviewed by anyone — since reviewing pull requests isn’t always fun or easy, and injecting in-depth human review slows everything down a lot — and so no one understands how the code works. (I keep thinking about the AI maximalist who recently said he woke up to 75 pull requests from his agent, like that was a good thing)
And maybe it’s a combination of the two: agent-generated pull requests are incrementally harder to grok, which makes reviewing more painful and take longer, which means more of them go without in-depth reviews.
But if your claim is true, the bottom line is that it means no one is fully reviewing code produced by agents.