Hacker Newsnew | past | comments | ask | show | jobs | submit | fsh's commentslogin

I would be very surprised if any of the frontier models wasn't trained on all public physics benchmarks. Training data providers have been hiring people for exactly this task.

This study appears to be evidence against that: the model failed the benchmark but arrived at the correct answer.

Half of the benchmarks don't have published answers. That's why training data providers have been hiring physicists to solve them.

This is unconventional benchmaxxing then, when they decrease the benchmark scores to allow models to generalize on correct solutions.

yeah, it would be almost shocking if an open source benchmark was NOT used ~somewhere in training. Perhaps just pre-training, but still. Neural networks can be fairly robust to some mistakes in their training data, so maybe it doesn't even matter if some of them are incorrect. Who knows.

I'd have thought the same but this article from yesterday blew my mind https://www.amazon.science/blog/why-dont-machine-learning-re...

As it essential implies that these models compress knowledge well in a way that what remains is what's generalizeable more so than remembering every specific detail... Anyway more understanding necessary but thought provoking


So after chasing AWS for two decades, they are still dwarved by them.

Launching far off the equator makes it more difficult and is therefore a bigger technological achievement.

The selection of launch site takes the orbital inclination into account.

https://en.wikipedia.org/wiki/Sun-synchronous_orbit

The "Onward and Upward" cube sats were deployed to Sun-synchronous orbits, with a roughly polar inclination.

Similarly, launches from USSF Base Vandenberg take a southerly trajectory, so it can be used for orbits like that.

Rocket launches from continental Europe have been challenging because of the population centers, and the unpopularity of things exploding over their heads, and falling on their houses and businesses. USA and Russia have some quite ideal sites that avoid flying over populated areas.


You are still ~500 m/s short compared to launching on an equatorial orbit from the equator.

Velocity is a vector, not simply a rate. Launches and orbits are three-dimensional.

What would be a great technological achievement would be launching from French Guiana (5.2° N) and then making a plane-change maneuver to achieve a polar orbit (90° inclination). All that bonus eastward velocity from the Earth's rotation would turn into a penalty they'd need to compensate for no reason, when they had a spaceport more suitable for that polar inclination.

Try this related concept: https://youtu.be/sod3ofdUYTw?si=U7dz_LcRQ0M44tAU

And the spherical cow jumps frictionlessly over the Moon!


Photonic quantum interfaces typically operate at glacial speeds and with terrible fidelities. There is a lot more missing than good qubits.

Electrical engineers in the 1940s already had "infinitely stable" memory in the form of delay lines and Williams tubes. Quantum error correction is incomparably more difficult than using digital bits.

Some cute tech projects do not justify a massive monopoly. Google is financing Waymo using the ~300 billion dollar it extracts annually from consumers via ad spending. I would rather pay substantially less on anything I buy, and have the free market sort out self-driving cars.


A product/service that can drastically reduce the top cause of injury and death, as well as give safe mobility to so many who do not currently have it is not a "cute tech project".

>I would rather pay substantially less on anything I buy, and have the free market sort out self-driving cars.

Alphabet's entire revenue divided by the total amount of goods and services sold is miniscule, so I don't see how it could be mathematically possible to pay substantially less on anything you buy.

>and have the free market sort out self-driving cars.

This is the free market sorting out self-driving cars.


Google is one of many monopolies that are collectively bleeding society dry. Of course, the others should be broken up as well.

Waymo is currently financed by a monopolist and has negligible revenue itself, so it is not part of the free market.


Self-driving cars are an absurdly inefficient way to reduce traffic deaths that exists only because it makes substantially more money for shareholders than the alternatives. You are being witheld a good and cheap solution now for the promise of a worse and expensive solution some time in the future. All the while people keep dying.

Oh, and what's this "those who don't have it" bullshit? If anything, turning driving into a subscription service will mean mobility becomes harder for many people, as dirt-cheap used cars slowly stop being available. As for people who can't drive for whatever reason, they already have taxi apps. And don't think for a second that self-driving taxis will be cheaper in the long term than regular ones, only the margins will go up.


I prefer to live in reality, where American suburbs are nowhere near dense enough for public transport to pencil out, plus there is no political will from the voters for it. No one is tearing up existing roads and replatting all the detached single family homes to more dense housing.

Waymo is a real solution to reducing injuries from distracted and drunk drivers, as well as reducing insurance costs. Obviously, if Waymo is the only game in town, then prices will be high but hopefully there will be competition. Owning a car 24/7 along with space to store it and insure it is not necessarily always cheaper than renting it per trip.

A robotic car can be utilized for far more hours per day amortizing its costs over far more trips making each trip cost less. Which, if there is competition, should result in lower prices.


> I prefer to live in reality, where American suburbs are nowhere near dense enough for public transport to pencil out, plus there is no political will from the voters for it. No one is tearing up existing roads and replatting all the detached single family homes to more dense housing.

Car Country was designed by the US car lobby. Political will of the voters? Oh yeah sure, now a century into it that’s also not palatable for home owners and the like.


> Oh, and what's this "those who don't have it" bullshit? If anything, turning driving into a subscription service will mean mobility becomes harder for many people, as dirt-cheap used cars slowly stop being available.

Too many people are in the "You Will Own Nothing And You Will Be Happy" camp.

Apparently the thought of lifelong indenture is a plus to these people :-/


* a good and cheap solution *

You mean public transport? I live in a country with one of the best public transport systems in the world. I enjoy it, but please don't fool yourself into thinking that it is cheap. It is not, very much NOT. All the big Czech cities spend between 25 and 35 per cent of their municipal budgets on public transport, and those are mostly already mature systems where the most expensive work like laying track is already amortized. New trams and buses, repairs, drivers wages, electricity, nothing is cheap nowadays.

Sure, an American city can decide to build a new tram or subway system from scratch, but don't promise anyone that it is going to be cheap. It won't, doubly so if the mayors stipulate high wages for the workers, like NYC does.

And even the best public transport system fails at the peripheries, where density of the population drops under a limit of viability. Then you have two buses a day even here in CZ.


Compared to every person owning their own car, but now it has a bunch of extra sensors and a cloud subscription? And the city still needs to build and maintain the roads.

Or a bunch of robotaxis, which reduce the up-front cost and amount of parking needed, but the operational and infrastructure costs remain.


[flagged]


free market :) (forgets about patents, tax cuts, excessive regulations thats unbearable for small players). Can it be that is not so free after all.


That's probably a lot more than their market share versus Claude Code and Codex.


I'm glad that my 9a has little enough RAM to be spared the worst of Google's on-device LLM features.


Shops used to curate their catalogs by in-house testing products from various suppliers. If a supplier comes up with a better toaster, they can send samples to the most relevant shops and convince them to stock it.


It's not VNC-based, so the performance is much better. It also supports multiple monitors.


Thanks!

GNOME Remote Login runs an RDP server, and RDP indeed isn’t fast enough, although is TLS-encrypted.


Isnt the issue with the latter is tha tyou need a local session first before you can connect?


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

Search: