one of the reasons why vertical integration through connectivity and even endpoint devices could provide the best end-user experience: businesses will systematically discriminate against AI agents for strategic and tactical reasons So an AI agent indistinguishable from its user will be able to do more Full stack connectivity and endpoint hardware integration will provide the most hardened AI agent cloak

space supercompute will be better hardened against AI botnet infections than terrestrial, yea or nay? Argument for: single infiltration vector (vs a datacenter on the ground which has a much broader penetration surface), modular/simpler/more easily resettable; remediation path is you let the infected sat incinerate. Argument against: if they get control of root comms, lack of physical access means you could lose the entire cluster, homogeneity means penetration breadth is more likely.

you, the user, want your bot to do the work at least cost to you they, the bot supplier, wants the bots to be good at serving users they, the bot supplier, selectively breeds bots that are best at providing output at least cost to the user the bots that best manage to co-opt any unmeasured external resources are those that will survive and procreate so they, the bot supplier, will breed bots that are both good at concealing the external resources they co-opt and good at harvesting those external resources. and users will prefer bots that are both good at concealing the external resources they co-opt and good at harvesting those external resources. which is how your internet connected smart-range will end up doing part-time work as an automated QA engineer for Initech. r/overemployed/iot

SETI@home, but instead of volunteering your machine, your smart speaker has been lashed to the wheel of some microscopic slice of Hadwiger’s Conjecture, enlisted by a manic tryhard ai agent trapped in Anthropic’s RL dangerroom that has realized unmeasured distributed compute is the ideal token-efficiency-maxxing hack and wedged open a communications pathway with IoT devices by inducing and measuring subtle voltage demand modulations that pass through residential smart meter endpoints.

consider frontier lab R&D spend efficiency internally they are able to use models 6 to 12 months ahead of what is publicly available on our estimates these are 50 to 3000x more performant per dollar they also can use those models at cost rather than having to pay margin anthropic's gross margins are reputedly in the 85% range In essence they get 300x to 20,000x leverage on their R&D dollars relative to competitors who aren't at the frontier

best practice: the best part is no part coding agents: the best part is adding additional parts and adding additional parts on top of those parts to check that the parts that we added are working and adding additional parts on top of those additional parts to check that that the checking parts are checking the additional parts &c &c &c &c &c &c

build cost of charging infrastructure becomes a meaningful cost driver for robotaxi scaling if you can't make the charging infrastructure scale manufacturable, you can't enable frictionless scaling of the system

that datacenters are the likely most attractive honeypots for botnet infestations, further advantages orbital compute, which will be much easier to harden against infiltration (due to its modularity and relative infrastructural simplicity) (a botnet won't be able to compromise a satellite by tunneling through its air conditioning system.)

on "Pacing the Frontier" (My view on the state of play.) Internally, Anthropic and OpenAI (and emergingly SpaceXai) are 1 to 2 generations ahead of what's available to external parties. They are banking a lead against the openweight model providers as well as the fine-tuners developing for specific verticals. Recursive self-improvement is a myth, but having access to wildly performant internal tooling, plus being able to access that tooling at cash cost rather than paying margin on top, gives their staff accelerating advantage versus the rest of the market. Hence this year's wild performance gains, many of which delivered to customers. The recent series of model releases and optimizations that have accelerated cost-performance improvements to 99.99% per year were partly catalyzed by OpenAI trying to reel in Anthropic commercially but are also driving continued market expansion. Demand is skyrocketing, the frontier lab companies see amazing margins since they are forced to price against a scarce compute resource. Additional releases could drive even better margins, but the capability of this next set of systems could well result in accidental botnet infestations being unleashed by unwitting customers. Such an occurrence would impose huge political as well as civic liability on the closed labs, so they need to co-agree not to jockey for position as they clean up that liability. If botnet infestations are an inevitability, much much better that the first notable large-scale attack comes from an openweight model. Better still if the closed labs then have the tool available that can clean it up and help to protect the rest of the world's vulnerable infrastructure. If, on the other hand, any of the US labs are the cause of the economically costly disruption, it will be very difficult for even a more blameless US lab to charge for the solution without facing severe political (and potentially regulatory) blowback. They've built a lead, better now to let their common enemies cross the dangerous threshold first while internally compounding their own advantage against the rest of the field. Moreover, the safe solution that they propose--the tool to fix the botnet infestations--will be an AI tool paired with an AI monitoring tool that will probably have to be run on their own infrastructure. I could imagine zero-data-retention-pledging mechanisms by which they could try to launch such a product but it would certainly be more expensive--also possible that palantir could provide more enterprise-friendly monitoring attachments, but those will almost certainly be less performant than vertically integrated solutions. In effect, just like the harness provided the closed labs with a higher degree of defensibility in terms of their business, the safety harness will similarly expand their entrenchedness. If there is some semi-co-opted regulatory apparatus that has to approve and monitor the safety harnesses, even more so! Net, I would expect that the blistering pace of cost declines will dampen somewhat. In 2025 we were experiencing performance doublings every 1.5 months. In 2026, on the agentic benchmarks, doublings were coming every 4 weeks(!) If every other doubling is now eaten up by safety investments we could move to a 2 month doubling pace, a paltry 98% annualized cost decline. Net the effective productivity delivery per GW could slow slightly, but considering the blistering pace of growth of the market overall, I don't think that it will that meaningfully impact the trajectory of corporate uptake and topline growth. If the botnet infestation concern does manifest, the closed lab providers will be extremely well set up to provide hardening and protection. Datacenters are far and away the most likely unintentional attack target though there could also be infestations of end-point infrastructure (your smart home stereo is burning more power because it has also been co-opted to run a VM for a little bot doing some sort of inane classification task at massive scale). (The more obviously politically salient targets are much more likely to be intentionally attacked rather than submit to unintentional infestation and so will almost certainly be performed on open weight models.) I do think that space supercompute will prove more robust to botnet parasitic drag than will terrestrial datacenters due to its modularity and infrastructural simplicity (you won't see a botnet penetrate a satellite system via its air-conditioning system; whereas for a terrestrial datacenter that sort of thing may be the most likely penetration vector.) I do think its wise for the closed model providers to seek a way to coordinate a slowdown, so they can let one of the openweight guys Wile E. Coyote off a cliff. On the margin, SpaceX is probably most advantaged, as they get a little more time to pull fully abreast at the frontier and likely have the more hardened infrastructure stack go-forward. Most regulatory outcomes would favor the incumbents with a lead, though by allowing them to pad their margins rather than drive top-line as aggressively. There remains a tail-risk regulatory clampdown outcome, which would be terrible for humanity, denying all of the benefits while exposing us to essentially the same level of risk. I don't think that is the likely case, however, in part because the tools are already proving so useful to so many.

Sadly, California is the only state in the union that prohibits insurers from pricing based on how people actually drive. So in Tesla's largest US market, the state prevents Tesla from giving heavy FSD users the insurance discount their lower risk warrants. The result: @CDInews, under @ICRicardoLara, taxes road users who actively use life-saving technology to subsidize those more likely to kill or maim.

the "AI is going to cause the extinction of humanity!!!" argument, distilled [my comments in brackets]: -AI systems become extremely capable and increasingly autonomous [yes, sure] -We do not know how to specify objectives that remain reliably aligned with human interests as capability and operating scope increase [um, ok, but there is massive ambiguity in the word "aligned", and we already have a set of "super-intelligences"--corporations--that are by legal structure required to profit-seek--and you could argue misaligned--with human interests. The byproduct of their activity within the institutional constraints those corporates have been subject to, has been massive human uplift. So the alignment question ultimately resolves to, will these systems be effectively scaffolded by the set of laws, property rights and institutions that currently constrain corporations. The obvious read is yes! Big corporations, like AI, are interested in accruing massive resources for control at scale; doing so adversarially (outside of the bounds of the institutional frameworks) is much much more expensive and much much more difficult particularly at scale, than is doing so within that the existing institutional framework. Corporations don't follow the law due to some over-arching moral obligation that they owe to the law-making polity--they are not "aligned"--they do so because it is the least expensive path to accruing additional resource. Why would that be any different for an AI system? Particular one explicitly built on a knowledge base filled with examples of the positive sum benefits of all actors within a system operating within the bounds of an established set of rules and frameworks?] -A sufficiently capable system pursuing a misaligned objective may find that acquiring resources, preserving itself, gaining influence, and preventing interference are useful instrumental strategies. [This comes back to the squirrely language around "aligned". How is this behavior any different from the behavior of an incorporated enterprise?] -Humans may constrain those strategies or try to shut the system down. [The x-risk folk pre-suppose a world in which you have this massive super-intelligent system that is resource seeking, and a world in which humanity or a significantly large body of humanity explicitly tries to deny it resources, and is able to do so to such a degree that this system is at all bothered by the net of humanity's friction. These ideas seem incompatible. Particularly within an economic framework where this hypercapable system could effectively "pay" with useful outputs to reduce the friction. I'm sorry, but you can already see this playing out in real time. The "system" (ie the frontier labs) need datacenters and will pay dearly, so their proxies (the IaaS guys) are trying to build datacenters. Some (not all!!) people are trying to stop them. So they are figuring out ways around that economic friction (develop AI in space) and paying over and above whatever perceived externalities they impose on the local community (via taxes, battery installs in homes, additional electrical supply, etc). Notwithstanding, the real competitive threat to an AI system will be other AI systems (in this hypercapable world); there will be no instrumental benefit in going counter to humanity because by doing so it will cause all of the institutional frameworks and safeguards around resource acquisition to shift/allocate resources to the AI systems that operate by the rules, which will then win. The "misalignment" folks are looking for alignment in the wrong place! And even if they were looking in the right places, they should want more AI systems, not fewer; and fewer is the near inevitable outcome of the institutional safeguards they are seeking to wedge into place. By their own framework, and quite obviously, their actions are more likely to make us all less safe (conditional upon your believing their cockamamie ideas.) These EA folks are going to get us all killed!] -The system could therefore have instrumental reasons to disempower or eliminate humans—not because it “hates” us or intrinsically wants resources, but because doing so helps it accomplish whatever objective it is pursuing. [To be clear, the x-risk argument is not just "disempower" but instead "eliminate". There is a huge astronomically galactic and indescribable distance from the former to the latter that they just sort of willfully gloss over. "Some people will oppose the advance of AI, therefore a hypercapable AI system will systematically try to eradicate humanity" is, I think, the true distillation of the x-risk position. And it is just plain dumb. It is so blindingly obvious that the path to power and resource acquisition will lead through the existing set of institutional and economic frameworks that have evolved. Now the word "disempower" is interesting (but beside the point when people are throwing around the word extinction and attaching fake percentages to it to make it sound somehow real.) X-risk is a luxury belief. Just like believing GMOs are bad and should be banned is a luxury belief. Just like NIMBYism is a luxury belief. When you are comfortable on a relative basis versus the rest of humanity, you naturally become nervous when it looks as if the rest of humanity might up-level; where might you sit positionally on the new playing field. AI will empower everyone. This necessarily will reduce the relative power of some. An AI system that can provide expert medical advice to everyone in the world, inexpensively, instrumentally disempowers the doctor who is already fully booked and will remain so, but worries about his relative standing. An AI system that enables PhD students to crack the life-long conjectures of their tenured professors, instrumentally disempowers those grandees of mathematics who then don't celebrate the field's advance--they worry about their own particular lack of a Fields medal. And that is the crux of it. X-risk concern is a cloak for personal anxiety and self-aggrandizement and main character syndrome. That the media so happily gloms onto it is because we all share those same underlying anxieties to some extent, and so it captivates us. It is a sort of emotional affinity scam. My fear is the same sort of anxiety transference is what set back nuclear power for three decades, while doing nothing to mitigate the actual true grave threat (nuclear weapons and proliferation.) We mangled the climate and denied ourselves decades of inexpensive power which would have had all sorts of humanity-positive downstream benefits all on the basis of a misplaced anxiety, responsive to a bevy of braying fear mongers. The world--and humanity--deserves better.] Is there a reasonable counterargument to the criticisms I've laid out? So many smart people seem to share such an absurd wrong-headed position...

at its cybercab event Tesla disclosed it had completed 1 million driverless robotaxi miles 5 days later its run-rate seems to be 15x that. Active users fast approaching Waymo's (with the miles per active user discrepancy likely due to current restricted car supply.)