skew-aberration

I'm going to lock in $325B.

People lack the accounting know-how to talk about it correctly. But compute for inference is just not the main expense right now.

If they take on more funding for training and/or build out, then they have to share the profits with more people. The growing revenue is shared by more people. As more options vest, they have to share with more people. If they need to acquire external companies for particular capabilities, profit is shared. If they have to pay hand over fist for hyperscaler compute in 2027 - more profit shared.

So as their revenue grows rapidly, the fraction of that revenue which they've promised to other people is growing too - case in point, they operated at an $80B loss on paper this year because the unrealized future-profits promised to early investors is so valuable relative to current revenues. If they have to make more such deals to continue investment (e.g a low IPO price which require shelling even more shares to stay competitive), then they will continue to create vast value but less of that value will be captured by the current investors (the 'burning cash' may go to compute providers, outside investors, as discounts to enterprise and consumers, etc instead of to Anthropic).

I have seen most revenue projections are on the order of 2-3x, but I suspect about 4-5x.

IMHO profitability will be worse though. They will contain to take on more and more investments - training, R&D, and 'sales and marketing' for enterprise share. They will face more competition than ever before. The range and types of models will continue to diversify, with real-time models for robotics and computer use paving the way for new entrants.

The investor confidence to back this doubling down will be spurred by rapid progress, panic deregulation while old-guard corporate democrats still hold power in the US, and much stricter sanctions against Chinese competitors, their models, and their eventual domestic hardware. The tech world will look a lot more factional by that point, with pseudo mergers & acquisitions amongst the big companies beginning, unabated by a toothless FTC. Anthropic will probably get closer to Amazon, and we could see big moves amongst Meta, Microsoft, and/or Google too.

Ultimately, Anthropic and OpenAI will maintain their market leading positions through at least 2028/2029 but only after a protracted period of significant equity dilution due to the expensive acquisitions and large outside investments which will be necessary in these market conditions.

Chinese hardware and models will look like a serious threat by the end of 2027 but will be largely locked out of US/Western markets. At the point the equation will have changed drastically - credit issues in the West led by mortgage security collapses will shape the investment environment and 2028 election, and rapid developments in physical technology and manufacturing will take the spotlight away from intelligence itself.

Small gripe but I don't like 'declines to 3x'. 3x growth would mean repeating this year's record growth twice over, which is smaller relatively but an increase in absolute growth rate.

Yes, starship funding itself as a workhorse for lifting to LEO is a good thing for the US.

That is not the point I made though. The question is why money was spent on starship as a lunar vehicle - that being the main criticism I have seen in the media (e.g. SmarterEveryDay video). Of course that could turn out to be misplaced, but there's no point denying that that is the main criticism.

> SLS still gets $2.5B every year, around $30 billion total so far

Not relevant to the HLS contract either.

But a) it is more like $40B and b) we don't have full numbers yet to make this comparison apples to apples. SLS costs come from the shorter timeline, the expensive shortcuts like ultralightweight shell, H2 fuel, disposable design, the crew rating, and the profit paid to manufacturers. We still don't know the equivalent numbers for starship: how long the timeline will be, the cost of taking the same payload to TLI, if the heavy structure will ever work as a lunar lander, if the booster can be used with an H2 upper stage / lander alternative, if the fully reusable architecture will work for lunar missions (depends on refueling performance), and finally how much profit SpaceX will make on these government missions (which depends on the market).

If those problems are all solved, and SpaceX gives NASA a good price, then we could see starship emerging as the best value somewhere between 2030 and 2035. Solving them in time for Artemis IV 2028 seems very unlikely now though.

Very similar to passive radar. That's why I'm asking about what the legal distinction actually is. If I made my two MLAT stations phase coherent, would that make a difference?

The main negative coverage I've heard is actually that starship is clearly optimized for the commercial LEO market while it has received to date $3B (of $5B) of government funds intended for space exploration as part of a contract it may never deliver on (at least, as it was originally envisaged), delaying Artemis by years and costing the government at least a further $5B in funding of gateway projects which will also be delayed and require expensive re-engineering if/when they are ever deployed.

This could be seen as either straightforward corruption or a gamble on the Starship platform for future Mars exploration.

There are very sound economic reasons why rockets for Moon exploration like Saturn-V were built the way they were. Saturn-V was around $35-40B in today's money for 13 rockets and 6 landings (excluding lander, crew costs). It took 5.5 years from greenlight to first successful orbital flight (which was also its first flight). Starship will likely meet this target of 6 landings after about 20 years of development, $25B R&D (excluding crew/lander contracts), and possibly about $5B in launch costs. Asymptotic costs will of course be much lower, but launch costs do not dominate space exploration costs. The platform might be capable of Mars missions one day - but it's not clear how far out that goal is.

So multistatic here is meaning multiple phase coherent receivers? MLAT is not considered multistatic passive radar?

Phase coherence is ITAR? Or its use in passive radar is?

Do you know of any good books or sources about this viewpoint?

When I read about history of Europe/Old World in Roman and post Roman times, it's clear that large scale movements of whole groups of people is very common in the historical record.

I wonder why such views were ever resisted in prehistory, a time when such movements were simpler and their motivations more likely to arise?

It is interesting that you consider R&D, design, and engineering as something fundamentally different to manufacturing, rather than its heart and soul.

"People of the same trade seldom meet together, even for merriment and diversion, but the conversation ends in a conspiracy against the public, or in some contrivance to raise prices."

Adam Smith, The Wealth of Nations

Land ownership is the biggest business and the source of the biggest conspiracies.

It is not just housing but all land - the grocery store ups prices without increasing profit because every landowner in the supply chain (who they rent their stores from, their warehouses, the factories that process and package foods, the farms themselves) ups rent too.

This is a predictable outcome of commerce due to the limited supply of land and is described in some detail by Adam Smith and developed by later thinkers like Riccardo and George (who took it to its logical conclusion and inspired the board game Monopoly).

During the 20th century, economics was wholesale rewritten to blur the lines between land and capital so as to bury this awkward conversation.

The general recommendation of these early economists was taxing land - moderate in the case of Smith, up to 100% per George. This amounts to public ownership of part of the profits of land ownership.

Presumably the law will be written to account for advertised role != assigned role, but it can't account for gaming the system by never hiring exactly the advertised position.

You can try adapting to this reality in your methodology though. Divide up your project into clear cut and ambitious modules, with well-defined objectives to evaluate each part.

Codebase porting, implementing sophisticated algorithms, performance tuning, model training, reverse engineering, fuzz testing / benchmarking suites, refactor to support a new interface, etc, scale so much better than 'fix a then b. don't break c' style prompts.

Using Fabel to improve some non-AI ML models a while back, I had a similar result - it could implement perfectly any suggestion I made, but it could not work autonomously due to making very poor suggestions itself, or getting distracted by the wrong things, making dumb assumptions, failing to see the big picture, etc.

These seem pretty easy to solve problems though if I could just afford 100 or 1000 agents. Let them explore and compete and work out their errors for themselves.

At a practical level though, I do a lot more 'bonus' R&D work now by just keeping the agents running and testing ideas in the background.

You wouldn't be concerned about breaking backward compatibility with rust though, would you? A recompile would be needed anyway and the change would be detected and enforced. The rationale for the bad API here (two types in one variable) was that users were ignoring NULL - that's the part which wouldn't happen in rust.

You might be better qualified as a shareholder then

How do you know the difference? Constant current is almost certainly a PWM controlled buck/boost regulator. The flicker is attenuated by inductance, more or less depending on the design.

Does it even make a difference for non-heated parts? Potable water is delivered in PVC pipes and HDPE hoses these days.

I've read many of these stories over the years. I find it fascinating that well-meaning and seemingly quite intelligent people are so eager to exonerate the very-obviously-guilty whenever they have any redeeming features at all. Charisma, intelligence, or just a similar background to the befriender. They refuse to accept that liars, cheaters, thieves, murderers, fraudsters, etc are often normal people like themselves who may even have admirable traits and qualities. Often, they do victimize people they truly love.

I believe this phenomenon is also rooted in the persecution fantasies that many people entertain, which play out in politics daily in all quarters. Like a kind of primal social anxiety - people have irrational fear of the crowd turning against without them having done anything wrong.

The undated and anonymous tip is meaningless and incredible, the policeman returning from holiday a few days early for a brutal murder investigation sounds very normal, the lack of DNA on the thoroughly cleaned murder weapon is irrelevant, the notion of the murderer killing Dianne hours before the robbery and waiting for the rest of the town to wake up before leaving is absurd, his 'alibi' for 4:30-6:30 is that there was such a continuous stream of customers to serve - yet not so many that any observed the robbery at 8:30 or saw the burglar enter hours earlier, or heard the brutal murder, and is anyway dependent on a theory that she was alive at 6:30 which presupposes the time of death was wrong, and therefore creates the opportunity for Robin to have also murdered her after 6:30.

Also, when a jury finds that a defendant probably or very likely committed a crime they must find them not guilty. A guilty verdict is reserved for proof beyond a reasonable doubt (not beyond all doubt though - an impossible standard). The minority jurors did not necessarily consider him innocent.

So they can avoid the embarrassment when Copilot migrates to Qwen 4

How do you prune out the 'fails to prune out behaviors' behavior? If my model is failing to prune it out, can another model prune out that behavior?

On the other hand, symbolic reasoning easily gets superhuman results in narrow domains. The Robbins Conjecture was proved on hardware like that in 1996 after 60 years. You could check out this theorem prover benchmark suite that has been maintained for over 30 years https://www.tptp.org/

I grew up with these machines around, and as a kid they seemed to so much more powerful than humans. It's kinda weird to think that I became 'smarter' than them eventually. I suspect they will be able to run superhuman AI one day too.

What's your setup? Flash in opencode requires far more oversight for me than Opus 5.5 in claude code.

> I don't want to relitigate Malthusianism

But you do want to assert that it is true, without being aware of its meaning?

Malthusianism is the notion that some people must be poor / hungry because humans choose to reproduce until they reach the limits of the environment to provide for them (especially those Irish potato farmers, according to Malthus), and that increases in productivity will not change this - they will just be absorbed by a growing population (which ought to be 'swept away').

In reality, growing populations have led to more than adequate improvements in food production efficiency and people are choosing to reproduce less rather than more.

There is an excellent summary and rebuttal of Malthusianism in the book 'Progress and Poverty' by Henry George, which offers an alternative view of how supply and demand for finite land (your words) rather than reproductive instinct (Malthusianism) creates poverty.

Deepseek 4.1 flash is so fast and cheap. For coding tasks, it only seems to lack the robustness of Opus 5.5 rather than 'intelligence' - it fails to double check critical assumptions, stops before the task is over, and doesn't do enough speculation or exploration on its own initiative. Yet it can do all these things when prompted.

There seems to be another magic ingredient for improving agents beyond model size & efficiency, and so far only OpenAI, Anthropic, and maybe Google know what it is.

It's ambiguous, because it lets the reader think for themselves about the implications of a straightforward fact?

If you do insist on being spoon fed - getting information about labor relations from Fortune magazine is like taking heath advice from a Luck Strike promotion. The article's cheery conclusion is: Yes, AI productivity is driving wages down - but we only need to be patient and wait for more jobs to replace them, not tax or regulate AI. Yes, unemployment is driving wages down - but more immigration will fix this, not make it worse. Yes, the deficit is high while corporate profits are up - but the solution is more jobs (how?) rather than more tax. Yes, smoking causes 'throat irritation' - but Lucky Strike is toasted.

I know nothing about fentanyl honestly, but my impression from heroin users at least is that it really is that good, and issues like addiction and withdrawal are somewhat overblown - many occasional users completely avoid them.

This seems obvious in current year, but extrapolating it I wonder where it takes us.

If society isn't there to keep us safe, free, and happy - what's the point? Accumulating lean certificates and energy production capacity?