So if there is some fringe element who's willing to use AI as a terrorist implement, they'll have very little competition and they'll be very little knowledge about how to counter them if we try to outlaw open source models. I don't really believe that we can make this tool any worse than any other tool in history that's capable of killing people in some way if it's misused, but simply driving it underground won't make it go away. It'll just put normal people who would be intellectually curious in jail for doing things that criminals are going to keep doing and exploit us with.
I think I've read this novel before, except it was rats and monkeys shocking themselves with electrically stimulated orgasms or something till they starved. Perhaps different hormones are involved, or it's remarkable that flies are able to delay gratification, but I'm becoming less surprised by this kind of thing as time goes on. We humans aren't as special as we thought, even when compared to flies.
Tell PA to stop granting them a local monopoly in my area then. I'll gladly switch when I have an alternative that works. I have nothing here in rural western PA except comcast and they have just been jacking the cost up year after year and sucking more and more.
Only idiots are basing their excitement about what's possible on those test scores. They're just an attempt to measure one bot against another. There is a strong possibility that they are only measuring how well the bot takes the test, and nothing at all about what the tests themselves purport to measure. I mean, those tests are probably similar to stuff that's in the training data.
Antimatter was to be used to generate the energy required, by annihilation of regular matte, not the antigravity itself. "Exotic matter" is undefined as of yet in that it could still exist as something else. Some people are confusing the two ideas, even though some (few) people probably did actually think anywhere would anti-gravitate, as it were.
I'm glad they settled this one; however, to my knowledge nobody thought Anti-Matter meant "Anti-gravity" w/rt Alcubierre warp drive. That's not what they meant by 'exotic matter.' But ok. sure. There was doubt that exotic matter could exist at all, and nobody had doubts that anti-matter could exist. I think there is something lost in the translation here. (No I'm not saying that warp drive still must be possible, I just think people, including the article writers, are conflating things.)
When you have millions of users, and your product has inherent danger, you can't assume the liability for all of them. (hammers and nails anyone?) The only reason that Microsoft has agreed to be liable for their users copyright stuff is because they know this case is a winner for OpenAI and that it does indeed meet merit for fair use. They wouldn't do that to 'be nice,' because not even microsoft can foot the bill for millions of users being sued. Their only alternative would be to not produce the product.
I think they successfully make their case and cite enough previous examples that they will win this. Individual users will get sued for infringement if they do silly things with the output, but I don't think training will be considered infringement after reading this document.
'Covers' require the payment of ASCAP fees -- usually done by the facility, not the performer, but if the facility does not pay, the performer can be liable. You even need a license to put a jukebox in your bar, so that argument doesn't hold up.
I worked with a team that was convinced of the superiority of elixir/erlang over ruby/rails in our ecosystem. It can accept thousands of simultaneous connections, etc. But the devil is in the details. If those connections include queries to a database then it's just a slow and clunky as every other front end framework and the learning curve is pretty damn steep. So overall the lack of integration with existing tooling and communities is what drove us away. Sure, you can do magnificent things if you keep every little bit in process but it's really hard to do that. And you need a super big box to make sure that everything is in one process.
Yes it would have. Planck mass black holes aren't even mentioned in this paper, I'm confused as to where this came from. PBH = 'primordial black hole' meaning formed at the beginning of the universe, not Planck Mass Black Hole, which, you're right, would have evaporated instantly.
I'm not a business owner or a CEO, and I can say wholeheartedly that I do not believe unions are good for anybody. It's just like rent control that gives renters lower prices but it destroys the supply eventually. Unions that get wage hikes by stomping their feet and yelling in public are not making the economy better. The detriment is that over time, it makes it harder to hire new workers because the union will get all up in your business if you try to fire someone for non performance. Union shops also make it harder for individual workers who stand out to negotiate higher salaries for themselves. On balance I think unions are a big bad.
They don't provide a fungible service. There's nobody else providing the same service. It's ok to be unhappy about a service you're getting and still using the service, especially when there's no alternative. TikTok is certainly no help.
Yup, that was also my take. I used to write 'on box' and you communicated between components using Com+, and when you switched to http, there was no reason to stick with just VB6.
This is crazy cool! Esolangs have been a hobby of mine, (more just an interest lately, since I haven't built any in a while,) so this is like a fun code golf game for compilation. Nice work, and even better, nice explanation article!
I was officially employed as a help-desk tech while I was going to school for CS. I wrote some internal-only software we used to track all of the installed software on all the hosts in our network and catalog them to a database. This was in 2000, so there weren't very many software packages that you could run like that would crawl the entire org and catalog the software installed without installing an agent on people's systems. This worked that way, and my VP came and asked me to help a team package it up so we could sell it as another product from my company. Looking back, what I wrote was crap, but boy was that a boost in the breeches, if you know what I mean.
True AI? What, like 'one model to rule them all?' Why does it matter how many models we are using? Do you have more than one computer in your computer? Is it a true computer?
I have often wondered about this: does padding and chopping to context size along arbitrary boundaries have lasting effect on the overall data (due to misalignment at the ends, some words or lines getting chopped in the middle, etc.) Does that net out, because we do it consistently, or do we lose vital information that's compounded across the entire training data set?
I don't understand why anyone is acting like the emperor has no clothes just because the 'program' they had supposed was one thing is actually several modules of things. How is this different from literally any other software project? This isn't The Highlander. There can be more than one.
This talks about two bodies, but would the Halo Drive (https://arxiv.org/abs/1903.03423) also be one such thing? I mean, I guess the photons are the other bodies?