It didn’t create a level playing field, it just discouraged a very specific type of R&D while ignoring all others. All other types of employee salaries follow certain rules and some can optionally follow R&D rules. Software is now the only one required to follow 5 year R&D amortization so the deck is now stacked against software.
The existing system is already corrupt so I don’t think that hypothetical is helpful here. The process by which contracts are awarded is already unfair and biased towards certain players with the right connections to get a contract written in a way that only they can fulfill.
It’s possible, although I have not verified, that the contract was written in such a way that Verizon was the only qualified applicant because of lobbying.
Sidestepping the corruption piece, which is a huge problem here to be clear, my understanding is that SpaceX has promised to provide starlink antennas at no cost to be used in commercial aircraft. Assuming that is true (again, a big if), then wouldn’t a natural consequence here be Starlink back Internet on the planes?
That said, I may have been reading misinformation and it’s only provided to ATC towers.
Bans from planes are already arbitrary and capricious, so I don’t think changing which arbitrary power is responsible really makes a difference here. Look at all the news articles of people ending up on the no fly list for garbage grounds.
> They could do a shitty job and keep the contract, so why would they bother to do well?
I think this is probably the key piece I didn’t consider. SpaceX’s previous work was done under heavy competitive pressure so they had to do good. Here it doesn’t matter what they do.
With Verizon, my doubts came from all of the reporting that’s happened on rural broadband funding. A lot of the companies, including Verizon, essentially took the money and ran without doing all of the work (from my understanding).
While I don’t like the way it is handled, I’m going to be a contrarian here and say this could be a good thing. Previously, the contract was awarded to Verizon at the same price. I have more faith in SpaceX’s ability to deliver than Verizon and if starlink antennas are on every plane, aircraft internet will likely be better overall as a side effect.
Please convince me why this view is incorrect, I am open to listen.
I’d argue digital ads kill people just via indirect means. Think of all the “dangerous” ads - fast food, alcohol, weight loss and plastic surgery, unnecessary medications, even guns. All of those things kill people over different time horizons and circumstances. Would individual people have bought the stuff without ads? Maybe, maybe not, but on a statistical level I’d guess it would be significantly less. So arguably ads do kill people and probably more than Lockheed Martin’s weapons do each year.
In the USA it is public record which in practice means anyone with money can get the record. This is potentially a large part of the high US recidivism. Once you have been convicted once, most employers will see the record and refuse to hire you forever.
Most natural gas in the U.S. comes as a byproduct of fracking oil, which imo is worse than coal as reserve power because you have to maintain fracking sites and usually pump a fair amount of oil and briny water alongside the gas. Coal is easier to mine in small amounts afaik and has less local harms (e.g. water contamination, earthquake risk, etc.).
Totally fair and I agree. But what about between now and eventually?
Eventually renewables will be all we use and eventually fossil fuels will no longer be needed. But between now and eventually, maintaining backup capacity is necessary and coal is probably the best option for that for the continental U.S. Nuclear only works as a base load, fracking/oil has even worse side effects, fusion isn’t ready, and we don’t have much untapped hydrothermal/geothermal
I think the counter argument is that renewables are never going to replace coal/oil/gas completely as there will always be the boogey man of “what if there is no wind/sun”. Having a small amount of fossil fuel based capacity in reserve would make a huge difference politically and of the options, coal is probably the best for that.
It is less environmentally damaging than maintaining fracking operations for oil/nat gas, extremely abundant in the U.S., and can be spun up or down on the order of hours so emissions can be kept minimal when plants are not needed.
And you probably could have said the same thing about Google 25 years ago when altavista and yahoo were around. What ended up happening was a complete dominance on web search for Google. The end result is unknown but it does appear they have an advantage right now.
At a purely technical level, no, as long as the model can output a null token. E.g. imagine training using a transcript of two people talking. What would be a single text token is a tuple of two tokens, one per person. Each segment where a person is not talking is a series of null tokens, one per ‘tick’ of time. In an actual conversation, one token in the tuple is user input and one is GPT prediction. Just disregard the user half of the tuple when determining whether the GPT should ‘speak’.
The real world challenge is threefold. First, null tokens would be massively over represented in training and by extent, in outputs. Second, at a computational level, outputting a continuous stream of tokens would be absurdly expensive. Third, there is not nearly as much training data of interspersed conversations as of monologues (e.g. research papers, this comment, etc.).