Deterministic responses are not expected from something clearly labeled able to make mistakes.

it's not clearly labeled. it's a search engine. i expect deterministic responses from that.

We already do. Libel and fraud are already illegal.

…even unintentionally.

Yes, obviously.

What if they just make a mistake, unintentionally?

Publishing lies that were generated by the plausible-lie-generating-machine is a very intentional action, it's not a "mistake".

The legal profession uses the term "reckless disregard", see, e.g., [1].

[1] https://en.wikipedia.org/wiki/New_York_Times_Co._v._Sullivan

If they do so knowingly, and harm is caused, then yes. Are you suggesting we should give them a pass from years of acquired jurisprudence simply because they hold a particular title?

And what institution gives out the licenses for journalists and scientists? Is it revokable?

Do LLMs knowingly make mistakes?

Do we sue scientists for unintentional procedural errors?

This isn't about what the LLM knows. No one's trying to hold a computer liable.

This is about what Google knows.

And Google knows perfectly well that LLMs hallucinate all the time. They will provide incorrect information, confidently and often.

Which is why this whole article is explicitly about holding Google accountable.

It’s in the nature of empirical reality that scientific measurements can be wrong.

Comcast is not a rapidly growing company, though.

This isn’t hypothetical. SpaceX is increasing Starlink revenue by like 50% per year. And their current Starlink constellation, about 10,000 satellites, has been launched entirely by Falcon 9. They’ve been waiting to launch much larger satellites on starship (in fact they had versions of these ready for several years now, and recently did suborbital tests of some of them on recent starship flights). Starship is about 5-10 times the capacity of Falcon 9, is fully reusable, & has larger diameter allowing much larger satellites. They asked for approval for roughly 40,000 of these larger satellites (~3 times the size of current ones, each about 10 times the bandwidth… and half of the 10,000 are even older designs), and they may eventually do about 100,000 of them & further increase the size and reduce latency (by operating at lower altitude). It’s not an exaggeration to say SpaceX intends to increase bandwidth by at least 100x, maybe a lot more. They intend to use a lot of this extra capacity to expand into mobile coverage as well. They are leveraging their platform for incredibly important national defense capabilities as well, and they operate as their own backhaul using on-board laser links. They can service anywhere in the world that will let them, including lucrative sectors like aviation. I do think it makes sense to value SpaceX as a rapidly growing business, not as a dividend-giving, plateaued ISP like Comcast.

This is all before even mentioning the idea of orbital datacenters.

You're trying to justify SpaceX's valuation using Starlink, but it's clear from the S-1 that it makes up only for ~5% of SpaceX estimated TAM.

Many people think their claimed TAM is total fiction, and attempting an actual realistic TAM relies far more heavily on starlink. From morningstar:

> Our base-case forecast entails $56 billion in revenue for Starlink in these niche and growth areas by 2035, representing about 45% of the identifiable market we’ve sized

source: https://www.morningstar.com/stocks/spacex-what-investors-nee...

> Many people think their claimed TAM is total fiction, and attempting an actual realistic TAM relies far more heavily on starlink.

Then either the TAM for Starlink is ~20x bigger than reported by SpaceX (I doubt they would downplay themselves in such a way) or the whole SpaceX TAM is ~5x smaller (much more realistic, if not more than that)

I think the TAM for both is reasonable (it's just generic "Enterprise" stuff, at the end of the day), but I think Starlink is better able to capture a larger portion of its TAM.

That is Tesla.

For now. SpaceX will be acquiring Tesla as soon as Elon gets around to it.

It’s only circular if it violates my narrative, of course.

It's circular when money flows from A to B and then back from B to A again.

This is a series of transactions in which money flows from Google to SpaceX. There is no flow of money from SpaceX to Google. So it's not a circle.

More to the point, allegations of circular financing are about following cash. When NVIDIA invests in a company so they can buy NVIDIA chips, that raises a unique set of questions distinct from other types of conflict of interest. Affiliated parties doing business, as is the case between Google and SpaceX, has its own host of conflict-of-interest concerns. But they're distinct from those that arise in circular transactions.

Or at least they are selling the idea of orbital data centers since the technology for orbital data centers does not exist yet or in the short term.

It’s literally just a satellite with GPUs on it. SpaceX already has about 100MW of satellites in orbit. They’re making an upgraded version of these satellites, V3, designed for launch on Starship. Making a bigger versions of those with larger radiators (because of the duty cycle) and putting them in sun-synch is all that is required, if it can be done cheaply enough. And Starship, if it works, should be cheap enough as its propellant cost would be less than the fuel cost of using airliners to do transpacific air freight.

Ok… but thats not a data center?

I’ve ran a small datacenter. 30kW can be a “datacenter”

Orbital datacenters is just an excuse of a reason to merge the two companies when there’s nothing else tying them together. They won’t actually happen.

SpaceX will run out of things to profitably launch with Starship once they launch Starlink V3 (a few hundred launches per year), and orbital datacenters provides effectively unlimited launch demand. Like Starlink was for Falcon 9, orbital datacenters are a way to fully leverage the insane launch cost advantage and launch capacity that Starship brings to the table.

I see your point. All this launch capacity is coming online but, once Starlink is built out, who is going to buy it? Surely Starlink maintenance isn't big enough [edit: maybe it is, it's hard to get my head around the number of sats]

The datacenters in space thing just doesn't make sense to me. Datacenters get their value from scale which is why they're so freaking massive here on earth. I just don't see any datacenter in space being big enough (unless built over multiple lifetimes) to use let alone profitable/desireable. What am i missing?

It didn't already happen. As you pointed out, people who funded the purchase of Twitter hold SpaceX shares, and this IPO is how they get their money back.

It did. After the merger with SpaceX, the Xai debt got renegotiated at a very low rate.

Starlink satellites can likely have their life extended by quite a bit, like past constellation satellites did. But it doesn’t really matter as SpaceX is still upgrading capability like crazy. Starship will mean a factor of 10-100x Starlink capacity expansion.

They most certainly cannot - LEO satellites literally fall out of the sky after about 5 years; they talk about this in the S-1.

It’s a function of altitude. They keep pushing the altitude lower. This is not the only option, although it has a lot of advantages.

> Ask a human a question like this

Why would a human know specs for a random phone off the top of their head? The human response is either "I don't know" or "let me look that up", not a hallucination.

I think that it feels a little wasteful to go to Google search to ask a question like this, only for the AI that's giving you an answer instead of page results to perform its own web search to get you the response.

Also, I asked a thinking model with browsing enabled and got this:

> The Google Pixel 10 is expected to support Wi-Fi 7 (802.11be), based on the Qualcomm Snapdragon 8 Gen 4 / Tensor G5 chipset it will likely use, which includes an integrated Wi-Fi 7 modem. Specific finalized specs aren't confirmed until Google's official announcement.

(Model GLM-5-Turbo - two months old - using Kilo Code in the "Ask" profile; in its thinking token churn it reasoned that it should keep the response brief and direct. Perhaps not the best suite of model+harness for this task, but it's what I had to hand that's not quantized to shit, is a thinking model, and has a web search tool available to it.)

> Ask a human a question like this, and they also have a chance of getting it wrong, even when confident.

We google something specifically because the humans within reach don't know. The goal of searching is, well, to search pages - we're trying to find a site when we use google search.

The goal when using an LLM is generally different; we want an answer, not a site.

LLMs can not point you to sites, only in a general direction. That is because complete URLs do not exist as single tokens in any of the large models. It can synthesize a plausible-looking url, and if you're lucky that URL might even exist. But that doesn't mean that there is any relation between between the text surrounding a hyperlink in LLM output and the text on the linked page.

AI agents can verify and summarize URLs, but a plain LLM can not.

I bow to your correction. I was using LLMs as a sloppy shorthand for modern AI agents with best interfaces.

*so long as an accurate answer exists on the internet

Claude is OK at saying when it can’t find good information, but it’s still 50/50 on citing a source that has nothing to do with its claim.

I think we’re getting what we deserve by snarkily telling people to Google stuff instead of answering accurately. Google results have never ever been pure accuracy

The point of LMGTFY is to land people on either the official documentation or a curated site like Stack Overflow. Google used to be able to do that reliably.

With the power of LLMs you can Google a standard library function and get an inaccurate summarisation of a Reddit discussion where neither side knows what they're talking about

Stack Overflow and Reddit for years have told people to just Google it. And then the Google result is people saying to just Google it, instead of actually being helpful.

Does this ban the prediction markets that don’t use real money but instead tokens that are worthless?

If the tokens are worthless, where's the thrill in the gambling?

Just for love of the game.

Then what is the need for tokens? They must have /some/ value, even if its bragging rights or something. Otherwise I can just get the same thrill from guessing if there will be a rainbow tomorrow and seeing if I am right.