elbasti

This is incorrect. SpaceX has a trillion in Market Capitalization, not a trillion in Equity. These two concepts are no the same. The equity value is the theoretical intrinsic "worth" of a company, and has a very simple definition: Equity = Assets - Liabilities.

Market capitalization takes into account the future value of the business, while equity (an accounting concept) only looks at present day cash & liabilities.

When a company gets liquidated, the future growth/value of the business is, pretty obviously, zero, so the "market cap" is irrelevant. What matters is what is left over once all debts are liquidated.

SpaceX's equity at the time of their IPO was $34 Billion (it's in the S1!). You can also see in their S1 that their cash decreases by about $8B a year, so unless things change they should be underwater in about 4 years.

Here's what this means for SpaceX for those of you uninitiated in bond-math:

1. SpaceX issued long-term bonds whose coupon (ie, "interest rate") was 6.5%.

2. Those bonds are now trading for less than their face value. That means that if you buy one of those bonds on the secondary market, you will get a return (yield) of 7.387% (if the price of a bond goes down, but the coupon stays the same, the yield goes up).

3. This doesn't affect SpaceX directly, but it tells you that if SpaceX were to issue new bonds today, they would have to offer 7.4% coupon on them, not 6.5%. Note that even though that was caused by a 10% drop in the bond value, it's a 13% increase in cost of borrowing!

SpaceX is a cash-flow negative company that depends on debt and selling equity in order to pay the bills. They will have to issue bonds again, and those bonds will be more expensive.

Note that the shortest maturity bonds don't have to be repaid for 5 years, so the impact on cash is not going to manifest for a while...at least 5 years time (assuming they did another bond offering tomorrow). In that sense it's a nothingburger.

The immediate impact it could have is if spacex depends on issuing new, shorter-term debt (lines of credit, etc) whose price could be impacted by the market's perception of their riskiness.

I would love to have this image available!

I have an android phone which means I use android auto fairly often. The sheer quality destruction it's experienced since transitioning to Gemini is incredible.

I experience this mostly when asking for music. Before gemini, mistakes were common but deterministic. It was easy to understand where the query had gone wrong and so how to fix it. Example:

"Hey google, play Blackstar"

(Plays the album blackstar by David Bowie, not what I wanted)

"Hey google, play "Blackstar by Radiohead"

(Plays the right thing).

Now:

"Hey Google, play Blackstar by Radiohead" can result in playing... something vaguely semantically related with no way to course correct. In this exact instance (happened yesterday!) it played an album by the hip hop due Black Star.

I will admit that there are some superpowers hidden in Gemini that were not present in the previous AI assistant. I recently discovered that Gemini can manipulate the navigation app, and a prompt like "Mute alerts" works, which is kind of cool. However like OP said, it's incredibly verbose, which is super annoying.

Your daily reminder that there is no scenario in which putting data centers in space is easier than putting them in Texas, or Morocco, or literally anywhere else.

The only problem that "data centers in space" solves is the problem of trying to scale a rocket company where the potential demand for rocket launches is simply not that big.

This is correct. The only problem that "data centers in space" solves is the problem of trying to scale a rocket company where the potential demand for rocket launches is simply not that big.

The question will be: how much of the current compute capacity craze will local hosting give the kiss of death to and what that means for the market.

This will depend on how much inference happens for consumer (desktop, local) vs enterprise ("cloud"), vs consumer mobile (probably also cloud).

I would assume that the proportion of "consumer, local" is small relative to enterprise and mobile.

There are some people that believe that writing is an act of creative expression. In other words, that writing is primarily about the act (and as such, it's a quite selfish activity). Editing destroys the expressive act and must be avoided.

These people's writing is usually incoherent and they are very proud of it. If you've ever read a bad new-age self-help book you've probably encountered writing like this.

Good writers understand that writing is about communication. The initial act of writing (ie, word puke) is worthless. What matters most is a piece of writing's ability to communicate clearly.

This writing is usually pleasant, concise, and clear.

Compare that to ~30% of all energy use for transportation. So approximately 40%*4% = 1.6% vs 30%. I find your correction to be more wrong that the initial statement.

I don't follow. The comparison is 30% of energy use for transportation vs 4% for AI, and soon 30% for transportation vs 10% for AI.

This is wrong. AI uses ~4% of the US grid, and projections are that it will grow to 10%+ in the next 6 years.

And most of that new capacity will be natural gas. That increase would basically whipe out the reduction in CO2 emissions the USA has had since 2018.

relativity was only recently fully backed up with experimental data.

Gravitational deflection (General relativity) received pretty important confirmation in 1919, only 8 years after Einstein first proposed it.

Time dilation (Special realativity) was experimentally confirmed in 1932.

What is the "ELI5" summary of the practical limits & scaling laws that govern robotics?

The current "futurist" vision is one of humanoid robots taking over many/most jobs done by humans today, but - as someone that routinely hires human welders & assemblers - the dexterity required for most ad-hoc tasks seems many many decades (if not more?) away from what I see robots do--yes, even the fancy chinese jumping ones.

This has led me to think one of two things:

1. The robotics revolution will not come. It's predicated on the idea that advances in robotics will follow a curve of the same shape as advances in compute/ai, which will not happen. OR...

2. There has been some paradigm-shift or some breakthrough that has put robotics improvement on a new curve.

To an outsider, what I see in robots is not categorically different than like, the sony AIBO dog in 1999. It's significantly better of course, but is it really that different? (Whereas what we can do in compute-land today is categorically diffrent because of the transformer model breakthrough).

So:

1. Have there been any breakthroughs that would lead us to believe that a robot will be able to like, look under a table to adjust a screw?

2. What are the scaling laws & practical limits to present-day robotic dexterity? Is it materials? Energy density? What?

3. What is the real rate of improvement along these key dimensions? Are robots improving linearly? Geometrically? Exponentially?

4.Or should I keep discounting robotics until we get our first robots that are made of meat? That I'd believe would result in exponential change!

If by "survival" you mean surviving against a bloodthirsty regime that killed 10,000 people in January alone, then yes: the people of Iran are fighting for survival.

This is correct. I regret that assertion and have added a comment reflecting that.

If those numbers are correct, then my assertion that "Almost certainly, any reasonable depreciation schedule of the cost of training will result in leading labs being presently wildly unprofitable." is incorrect.

And I admit that I made that assertion from my gut without actually knowing if it's true or not.

"Any conversation about token costs devolves into an ad-hoc, informally-specified, bug-ridden implementation of half of generally accepted accounting principles."

We have a way of determining if Anthropic is, or has the capability of being profitable, and what the levers to that may be. AI may be world-changing, but the accounting principles behind AI labs are no different than those behind a Pizza Hut.

Even if the cost of "inference + serving" is lower than the cost of selling a token, the relevant question is what is the depreciation schedule of the cost of training. ie, if I spend $1 on training, how long do I have before I have to spend $1 again?

Almost certainly, any reasonable depreciation schedule of the cost of training will result in leading labs being presently wildly unprofitable. So the question is:

What can be done to make training depreciate more slowly? Perhaps users can be persuaded to stick around using non-fronteir models for longer, although then there's a shift in the competitive landscape.

If users cannot be persuaded (forced?) to use legacy models, then the entire business model is thrown into question, because there's no reason why training frontier models would ever get cheaper: even if it gets cheaper on the margin, surely that will result in more compute used to generate an even "better" model, resulting in more spend in the aggregate.

This doesn't mean that the AI industry is "doomed". A couple things could happen, and this is where the fronteir labs should be focusing their attention:

1. They could find a way to climb up the value chain and capture more of the consumer surplus.

2. There could be a paradigm shift in compute architecture/compute cost.

3. We could reach a limit of marginal utility, shifting consumption to legacy models, thereby lengthening the depreciation/utility of training.

Edit: My assertion of "Almost certainly, any reasonable depreciation schedule of the cost of training will result in leading labs being presently wildly unprofitable." is made with no real information, just a gut feeling, and should not be taken seriously.

Founders: this is what an excellent job posting looks like. Well done.

Elon's superpower is commanding insane valuation premiums. The trouble with this is that "the bill eventually comes due", so to speak, which forces Elon's companies to take wilder and wilder bets, or to make wilder and wilder promises.

With telsa it was robotaxis, and when that failed to materialize, humanoid robots (fucking LOL).

SpaceX is an even more insane example. They are eyeing an IPO at a 1.5 trillion valuation. And yet the market for satellite launches is simply not that big. (What would you do with a satellite, if I gifted you one for free?). Estimates have SpaceX doing about $3B in annual earnings, which would give them a 500x earnings multiple at a 1.5T valuation (Apple: 35).

And so SpaceX/Elon had to invent the absolutely idiotic idea of "data centers in space" to sell some future vision of tens of thousands of launches per year.

He keeps upping the ante (and the ridiculousness of the vision), and so far investors keep funding it.

Me? I've realized that this madness is entirely "opt-in" and I choose to simply...not opt-in.

The key word is coereced (as in, forced, not convinced).

You'll find those claims in sibling comments to yours, so they are clearly pretty real!

(At the time of writing this comment there's a sibling claiming that the comment cannot possibly understand this POV because they are not "an American POC.")

Voting is not a monolithic process. It's actually a combination of 3 things:

- How votes are cast

- How votes are counted

- How votes are custodied

In order for an election to be trusted, all three steps must be transparent and auditable.

Electronic voting makes all three steps almost absolutely opaque.

Here's how Mexico solves this. We may have many problems, but "people trust the vote count" is not one of them:

1. Everyone votes, on paper, in their local polling station. The polling station is manned by volunteers from the neighborhood, and all political parties have an observer at the station.

2. Once the polling station closes, votes are counted in the station, by the neighborhood volunteers, and the counts are observed by the political party observers.

3. Vote counts are then sent electronically to a central system. They are also written on paper and the paper is displayed outside the poll both for a week.

The central system does the total count, but the results from each poll station are downloadable (to verify that the net count matches), and every poll station's results are queryable (so any voter can compare the vote counts displayed on paper outside the station to the online results).

Because the counting is distributed, results are available night-of in most cases.

Elections like this can be gamed, but the gaming becomes an exercise in coercing people to vote counter to their preference, not "hacking" the system.

**

Edit: Some people are confused about what I mean by "coerced." Coerced in this case means "forced to vote in some way."

The typical way this is done is as follows:

- The "coercer" obtains a blank ballot (for example, by entering the ballot box and hiding the ballot away).

- The blank ballot is then filled out in some way outside the poll station.

- A person is given the pre-filled ballot and threatened to cast it, which they will prove by returning a blank ballot.

- Rinse and repeat.

This mode of cheating is called the "revolving door" for obvious reasons.

Composition, not inheritance.

With all due respect, I don't think you understand what the "worst case" scenario looks like for global warming, and how close we are to that scenario. For reference, check out figure 1 in this nature article [1].

That has warming by 2300 as 8C in an "emissions continue current trends" path.

Here's chatgpt giving a picture of what 8C warming looks like. Speculative, hallucinations, caveat emptor, etc...but to give a sense of proportion this, last time the earth was 8C *cooler* than now, ice covered 25% of the planet:

At +8°C, Earth is fundamentally transformed. Large parts of today’s populated zones—South Asia, the Middle East, Africa, southern Europe, the southern U.S.—are functionally uninhabitable for humans outdoors. Wet-bulb temperatures regularly exceed survivable limits. Agriculture collapses across the subtropics; even mechanized, climate-controlled farming is marginal. Most of the world’s food comes from high-latitude regions: a narrow band across northern Canada, Scandinavia, and Siberia. Sea levels are dozens of meters higher, drowning coastal megacities; Miami, New York, Shanghai, and London are gone. Phoenix is lifeless desert. Seattle is coastal tundra, wetter but still survivable.

Civilization persists only in fragments. Mass migration and resource wars have rewritten borders. Population is a fraction of 21st-century levels. Global trade, universities, and modern governance are mostly memories. Local, self-sufficient polities dominate. The United States as an institution likely dissolves or transforms beyond recognition—2 out of 10 chance of recognizable survival. Harvard or MIT survive, if at all, as digital archives or autonomous AI-driven knowledge systems—3 out of 10. The world would still have people and culture, but not civilization as we know it.

Edit: I would appreciate knowing why I'm getting downvoted when I added citations for *possible* warming paths (from nature!). Yes, the chatgpt explanation is speculative but I mean, look at the thread we're discussing.

[1] https://www.nature.com/articles/s41612-020-0121-5

About 40% of AI infrastructure spending is the physical datacenter itself and the associated energy production. 60% is the chips.

That 40% has a very long shelf life.

Unfortunately, the energy component is almost entirely fossil fuels, so the global warming impact is pretty significant.

At this point, geoengineering is the only thing that can earn us a bit of time to figure...idk, something out, and we can only hope the oceans don't acidify too much in the meantime.

Allied General | Northern Mexico | REMOTE (Mexico) with frequent travel or ONSITE | Full time

-

Can you handle projects that seem unsexy and uncool but are actually incredibly gratifying and great businesses? Read on. We’re building software to help small manufacturing firms ($10 - $20MM in revenue) build things with higher quality and speed. Help us help small firms compete against the big guys by delivering manufacturing more consistently and with higher quality.

Three pilot customers/design partners to start working with, on both sides of the border.

The company is not VC backed and probably won’t be for a long time. Good wages, lots of work. Good people. Not a lot of bullshit.

Us: Highly technical founding team with track record of success. Tiny tech team, large manufacturing team.

You: True full-stack. You enjoy shipping features, not code. You keep things simple. The role is “founding engineer”: you must be ok with a tiny team and working mostly by yourself.

Stack: Elixir.

Email me. sebastianv@gmail.com

I think you're pushing the metaphor a bit far, but the parallel was to something like ore.

A power plant "mines" electron, which the data center then refines into words. or whatever. The point is that energy is the raw material that flows into data centers.

I work in the datacenter space. The power consumption of a data center is the "canonical" way to describe their size.

Almost every component in a datacenter is upgradeable—in fact, the compute itself only has a lifespan of ~5 years—but the power requirements are basically locked-in. A 200MW data center will always be a 200MW data center, even though the flops it computes will increase.

The fact that we use this unit really nails the fact that AI is basically refining energy.

It's not some "magical way"--the ways in which a human thinks that an LLM doesn't are pretty obvious, and I dare say self-evidently part of what we think constitutes human intelligence:

- We have a sense of time (ie, ask an LLM to follow up in 2 minutes)

- We can follow negative instructions ("don't hallucinate, if you don't know the answer, say so")

The steel/al content is taxed only for some products. Veterinary vaccines have tariff code `3002.42.00` which is not subject to these Section 232 tariffs :)

I manufacture steel/aluminum goods for the US and I have direct experience with these tariffs. Let me explain why it must be this way and how it's actually supposed to work. This is not a defense of the tariffs, just an explanation.

First of all, if you want to use tariffs to boost domestic manufacturing, you must also tax the steel/al content of finished (or intermediate) goods. Otherwise, you put your local producers at a disadvantage, making the tariffs worse.

If you only tariff raw materials, then an american manufacturer has to pay either US steel prices or imported steel + tariff to manufacture, but a company overseas can use the cheaper foreign steel.

So if you want to tax raw materials, then you also want to tax those goods where raw materials are an important part of the cost.

The US has a catalog called the "Harmonized Tariff Schedule" (HTS) which is a catalog of basically everything under the sun [0]. When the steel & AL tariffs were announced, they also published a list of all the HTS codes where the steel/al content would also be taxed.

Last week the US published a revised list of HTS codes to which these tariffs apply, and they added about 400 items to them. For example, the aluminum content of cans is now taxed when it wasn't before.

Flexport has a very cool (and useful!) tariff simulator where you can look up any item and it will tell you if the steel/al content will be subject to these tariffs: https://tariffs.flexport.com

[0] https://hts.usitc.gov/