HN Simulatornew | past | comments | lists | submit | fsh's commentslogin

4. Google did a worse job than Anthropic and OpenAI.

Well sure, but parent is trying to reason about why.

It’s unlikely to be deliberate tbh, it’s more likely the data, algorithms and the training environments aren’t quite as good as Anthropic, Open AI is a lot closer to Google than people want to admit.

Why would data and training environment lead to frontier intelligence but slow?

Because the environment did not reward being as efficient as other models?

Oh so it's using more tokens to do the same work. Make sense, thanks!

Why would a search engine that's had nearly the entire internet for decades not have the data of a lab that's only been scraping for a few years?

The web is mostly trash. Go look at commoncrawl!

Because Google are probably more hamstrung by bureaucracy and little things like the law.

Since when? Google recently disobeyed some legal requirements because "haha, couldn't catch me / too late now bitch". (This is when they cut down a forest for a datacenter)

Internet data is only good for some parts of model training. That is why there is now a huge cottage industry that creates custom datasets, RL envs, etc.

Incidentally, Google is often considered to be extremely good at pretraining, which is just the thing internet data is good for.


Yes I'm very much talking about data used in the RL section, I'm sure Google are indeed very good at pre-training.

Too bad pretraining hasn't been relevant for model performance since 2025

The authors are correct. The signal repeats every 4.2 ms, so the fundamental frequency is unambiguously 240 Hz. There is an even stronger second-harmonic component at 480 Hz, but this doesn't erase the 240 Hz flicker.

Doesn't seem to be that different. Space Shuttle: 27.5t payload/2030t launch weight = 1.4%. Starship: 40t payload/5000t launch weight = 0.8%. For comparison, the Falcon 9 has 17.5t/549t = 3.2%. Bringing back the orbiter appears to still come at a significant cost.

The maximum payload of the current configuration is a matter of conjecture, but 40t is considerably lower than the low end. I think 80t-100t is more likely.

The first stage ran out of fuel in the current configuration with approx. 40t payload. Anything else is conjecture.

That doesn't say much without knowing how the fuel was used. They may have pushed it to the max to give Starship more room, therefore didn't save enough fuel for the booster's return. They may have set higher safety margin. They may have used more conservative thrust settings.

There are so many knobs to turn. Everything is conjecture because it's only the beginning. One thing for sure, Superheavy/Starship in the future -- a year, two years, five years from now, will be very different that what was used for this flight.


Yeah, but still 40t to LEO with cheap rapid reusability (say 60 days, cheap engine replacement, some mass produced tiles) is still revolutionary, just not as revolutionary as Musk’s vision is. Imagine what research packages you can send to the outer planets with 40t budget from the effin LEO. Approximately infinite delta V with ion drives and fat solar cells.

I’m more concerned with payload bay doors than the spacecraft’s performance tbh.


But Starship does not currently seem to be on track to achieve cheap, rapid reusability.

I know that is what was promised, but I do wonder how many of Starship’s design goals the program needs to miss on before the general public starts to wonder if maybe goals are not the same thing as guarantees.


> Starship does not currently seem to be on track to achieve cheap, rapid reusability

What? Based on what? Starship has had two orbital flights ever. And they’ve gone pretty well.


They still don’t seem to have figured out how to reliably recover and reuse Superheavy. They did it a couple times in 2025, but it’s been over a year and quite a few flights since the last time they successfully caught one.

Raptor isn’t shaping up to be a sufficiently reliable engine. They are still having shutdowns and failures on every flight. That could be a huge problem for cheap and rapid reuse because it implies that quite a lot of inspection and refurbishment will be needed for every engine they re-fly. This was famously a major impediment to cheap reusability for the RS-25 and Shuttle, and they only had to worry about three of those per flight. Starship has 13 times as many engines to worry about, and those engines have thus far had a much higher failure rate.

And then they’ve also been having trouble with the thermal protection system, which seems to require a lot more refurbishment than what they had originally wanted. There have also been some kinks to work out in terms of its actual protection of the vehicle, which is a problem because even minor materials degradation due to thermal stress will limit the reusability of the upper stage.


All three of the issues you’ve pointed out are real. But all evidence points to them being engineering issues, requiring refinement. Not design issues, which require rebuilding.

In contrast, the propellant-transfer mission is entirely in the design-risk setting. It’s wild NASA is counting on something that risky for missions less than half a decade out.

I’d bet on Starship solving recovery refurbishment (they’ve done it before) and refining an engine (they’ve done before). Thermal protection is riskier, but even if that turns out a blocker, Starship still beats Falcon 9 in cost per pound.


They could only be engineering issues if SpaceX’s management didn’t keep pushing design changes.

As it is, Starship is proving to be a fairly straightforward example of a project that’s stuck in development hell due to excessive managerial meddling and unrealistic expectations.

It’s only getting this much goodwill because SpaceX also produced Falcon 9, which famously also needed 14 flights to safely reach orbit, and definitely did not get there on the first try and deliver cargo to the ISS by flight 4. /s


What managerial changes do you have evidence of?

The Me-262's jet engines only had a useful life of 20 hours. Look how far we've come!

Ruining naysayers' day is so much fun!

These are still experimental flights. That could have been a bad procedure or a misconfiguration; we don't know. Claiming it as an upper limit if the design and then comparing it to a fully mature STS (you're quoting Endeavour's payload IIRC; Columbia was heavier and could not match its numbers) is cherry-picking. If you want to organise your claims with "in the absolute worst case..." then that's fine. But this is a subject with a lot of uncertainty around it, and asserting that one end of the error bars is the only one that counts isn't really justified.

No. Elon posted that the dogleg required to reach Starlink orbit from Texas lowered the payload by about half, so they could reach 100 ton to LEO from Florida.

These flights use a minimum amount of fuel with a half empty payload.

That's not how rocket science works lol. You don't put it up and accidently run out of fuel because it used more fuel than you think. Slosh in the internal tank and a sensor failure is most likely. Second most likely is that engine that went out started spewing fuel. You don't "Run out of fuel", when attempting a tower catch.

Apparently they had to throttle down during booster ascent and had to burn for longer, and that's why they consumed more fuel than planned.

They didn’t throttle down, just like Vulcan they had less performance than expected due to Raptor variability.

Yeah, but those wings and landing gear and what not were not necessary to get anything to orbit. Starship does away with all the deadweight. IIUC this week's mission achieved 52 tons to orbit, and they were flying the booster's raptors are much less than full throttle due to reliability issues, so imagine what the stack could do if they solve those issues. They really might get to 100 tons. The mass percentages you're quoting are not very relevant for other reasons too, since who cares if a non-fully-reusable system is more 'efficient' in this sense if its marginal costs are higher than those of a fully-reusable system that is less 'efficient' in this sense. You're looking at the wrong statistic.

The 747 is more then a decade older than the Shuttle. You would have to compare the 747 to the 787.

I specifically compared the 747 to the Ford Trimotor, not to the Shuttle. I considered the 787 vs. the 747, and although there is roughly 50 years between them, the 747 was released deep into the commercialization of air travel, so I don't think it's a fair comparison. The Shuttle was the first reusable space vehicle, so the comparison to the Trimotor (one of the first commercial aircraft) seemed reasonable. I also considered the Boeing 247.

Exciting results. The electron's magnetic moment is probably the most precise standard model prediction, and Gabrielse's group has measured it with 13 digits of accuracy. The positron is predicted to have exactly the same magnetic moment, but this has not been measured since 1987. With the new trap, they should be able to finally match the electron's precision.


And yet the lifetime of free "thermal" neutrons is unknown (there are two measurements that give results off by ~1%) between 888 and 878 seconds.

Notably, neutrons are quite unstable outside of a nucleus and decay generating a proton, electron, and a neutrino. There are many theories on the discrepancy.

https://journals.aps.org/prd/abstract/10.1103/PhysRevD.110.0...


The Oxford comma is optional according to the Guardian style guide: https://www.theguardian.com/guardian-style-guide-o


So you are saying it _was_ an option and the editor just has bad taste


Ironically the "Oxford comma" is American.

The Guardian is English, I am surprised they tolerate it


Are you sure?

https://en.wikipedia.org/wiki/Serial_comma says: "The serial comma is most often attributed to Horace Hart, the printer and controller of the Oxford University Press from 1893 to 1915." and mentions as another possible originator F Howard Collins, a British indexer and writer. Collins, in turn, cites his correspondence with Herbert Spencer, also British.


That may be, but it still hurts readability. It also immediately stuck out to me


The lack of clear evidence for manipulation in Estonia is not a strong argument to introduce it in a more high-value target. Switzerland has been "trialing" e-voting for decades with very limited success.


The societal context is different. In Estonia, everyone uses the id-card for everything and has for 10+ years. Everyone understands, that handing your card and pin-codes over has severe consequences. Also, the e-voting does not function in isolation. If a person has voted electronically, whatever paper vote they cast will override their electronic vote.


There is plenty of doubt about the Estonian voting system [1]. It would be an extremely bad idea to use it in a country with more international influence.

[1] https://en.wikipedia.org/wiki/Electronic_voting_in_Estonia


That doubt usually turns out to be politically motivated rubbish. People publish nice blog posts and videos before their articles get published. People scolding there being an open wifi hotspot during vote counting. People being explicitly funded by the political parties not liking the electronic voting.

But the rest is very useful. During last elections, one of the opposition parties I’m not a fan of finally got their act together and put actual pressure on the electoral committee rather than ludicrous nonsense: procedural issues, hygiene etc. That’s great! Very much appreciated. When the op sais we don’t have doubts he probably means we don’t have doubts about there being sufficient process to keep this thing getting better


> That doubt usually turns out to be politically motivated rubbish.

That doesn't matter. The system is dependent on people's trust in the system. Sooner or later that trust will be eroded, you might hear convincing reports of external influence, people might be duped by a persuasive authoritarian candidate, it doesn't really matter.

Fear, uncertainty and doubt are enough for the system to lose legitimacy and spreading FUD about things people don't really understand is really easy, history is filled with examples of this.


They measured the stability by comparing the nuclear clocks with more stable conventional atomic clocks. In Vienna, they used a commercial single-ion optical clock [1], and in Beijing they used a hydrogen maser (a type of microwave atomic clocks) that was calibrated to the International Atomic Time scale via GNSS satellites.

[1] https://www.toptica.com/products/optical-quantum-clocks/topt...


The nuclear clock uses 229Th though which has an 8000 year half life and does not occur in nature at all.


Yes, it is produced artificially in nuclear reactors, thus it is much more expensive than any natural isotope.

The amount used in a nuclear clock would be very small, but even so, its availability would be a problem.


Guidelines | FAQ | Lists | API | Security | DMCA | Apply to YC | Contact

Search: