| ▲ | LastTrain 14 hours ago |
| On the one hand I completely understand and embrace that there is no stopping progress and adapt or die and all that. On the other hand I understand that we've built a machine that can strip-mine all of human knowledge for the benefit of a very small number of people and that sucks. |
|
| ▲ | 27m 13 hours ago | parent | next [-] |
| > we've built a machine that can strip-mine all of human knowledge for the benefit of a very small number of people Probably the best way I’ve heard anyone describe this |
| |
| ▲ | disqard 4 hours ago | parent [-] | | Indeed, well-stated! Another gem is: The true purpose of AI is to allow wealth to access skill without allowing skill to access wealth. |
|
|
| ▲ | tim333 an hour ago | parent | prev | next [-] |
| It's kind of for the benefit of everyone. Anyone can use AI, often for free or open source and the same knowledge is mostly there in all the large models. Information wants to be free as they say. |
| |
| ▲ | CalRobert an hour ago | parent [-] | | Anyone who is allowed to can use AI. Europeans were already locked out of some models for a bit. |
|
|
| ▲ | Palomides 14 hours ago | parent | prev | next [-] |
| ok but is it overpriced or underpriced? |
| |
| ▲ | beAbU 3 hours ago | parent | next [-] | | It is tremendously under priced! If you factor in the cost of all the source material used to train the models, with proper attribution and royalties and whatever, and you consider the massive cost of training, and that it has to be redone every couple of months for a new model, then I don't think any mortal will be able to afford any AI product. | |
| ▲ | LastTrain 14 hours ago | parent | prev [-] | | I don't know I can't decide if not actually owning any of your infrastructure & energy capacity is a good thing or a bad thing in this space. |
|
|
| ▲ | tamimio 10 hours ago | parent | prev | next [-] |
| It’s why it’s more important than ever to invest in open weights models and local AI. |
|
| ▲ | CamperBob2 13 hours ago | parent | prev [-] |
| (a) The stuff they 'strip-mined' is still there; (b) It benefits everyone who can afford a $20/month subscription. |
| |
| ▲ | steve_adams_86 13 hours ago | parent | next [-] | | (a) Kind of; aren't they buying and destroying incredible numbers of books at an alarming pace? You could say the data from the books still exists (digitally) but no one can access it other than them, now. Not to say we'd easily access most of it either, but that does still seem subtractive to me. | | |
| ▲ | CamperBob2 13 hours ago | parent [-] | | That's an artifact of our stupidly-broken copyright laws, and not Anthropic's fault in the slightest. If it bothers you, don't complain about AI, work for copyright reform. | | |
| ▲ | greggoB 12 hours ago | parent [-] | | You're going to have to spell it out for me how the physical destruction of books has anything to do with copyright. Google managed to scan books 20 or so years ago non-destructively, for reference [0]. [0] https://www.npr.org/sections/library/2009/04/the_granting_of... | | |
| ▲ | cowlevel 2 hours ago | parent | next [-] | | Google has an exception because it's a library. Anthropic is not a library. | |
| ▲ | positr0n 8 hours ago | parent | prev | next [-] | | Yes they were trying to do the right thing and provide free access to old books nobody wanted and then they got sued and stopped because of aforementioned copyright laws so now nobody has them. | |
| ▲ | CamperBob2 8 hours ago | parent | prev [-] | | Anthropic paid a large settlement -- over a billion dollars IIRC -- to copyright holders who sued them for training on their material. As I understand it (and you would certainly want to check with a knowledgeable source) their act of training on the material was subject to a valid fair-use defense, but the act of downloading it was not. Apparently it is considered OK to train on copyrighted material from a book as long as you own a legitimate copy of the book in question. So, having bought the copy necessary to be able to train on its contents legally, they can never sell or give away that copy without losing that right. Hence the need to destroy it. Yes, this is as stupid as it sounds, but again, that's my understanding. |
|
|
| |
| ▲ | emaro 4 hours ago | parent | prev | next [-] | | re (b): in absolute terms, yes. But does it hold in relative terms as well? How are you going to compete with your $20 plan with people who can affort the $200 plan or with OAI/ANT internals who are buring $1000s/day and have access to unreleased models? In comparsion, you're worse off than before. It's the same with wages: yes, they've gone up, so you should be better off. But most people loose anyway because a) costs have increased more than their salary and b) wealth concentration increased as well, so in comparsion they're also not better off. | |
| ▲ | LastTrain 13 hours ago | parent | prev | next [-] | | (a) yes and (b) yes the masses pay 20 a month to the few. that is what I was talking about. | | |
| ▲ | CamperBob2 13 hours ago | parent [-] | | And that $20/month buys me a lot of power. | | |
| ▲ | bpavuk 12 hours ago | parent | next [-] | | that they control and not you, re: tightening cyber safeguards. earlier I commented on "Gemini 1.5 through 2.5 being decent RE sidekicks for free before Google's rug pull." now I decided to throw a buck for API tokens, and I discovered that if I want to use a Gemini model even for semantic decompilation (aka dump control flow graph into a LLM, together with Ghidra output, and let it figure out a cleaner C/C++ pseudocode), I'm hit with "prohibited content" errors on Gemini 4 models and 3.8 Flash. | | |
| ▲ | CamperBob2 4 hours ago | parent [-] | | That's a problem, all right. I run those sorts of tasks on local models, which are (so far) relatively easy to uncensor. |
| |
| ▲ | LastTrain 12 hours ago | parent | prev [-] | | If it can be had for $20 it isn't any power at all, relatively speaking. There's nothing you can make for that $20 a month that they can't do a hundred million times over. No meaningful number of people are going to make money consuming AI. |
|
| |
| ▲ | Choad_20261104 12 hours ago | parent | prev [-] | | back a decade ago, when i was studying LLMs and generative models the only way to get a foothold was to 'give away' your expertise in public forums like github. When Microsoft bought it, I was certain I would never practice there and it hurt me because that was the ONLY way anyone's expertise was legitimized. Without github evidence, you couldn't get a word in. I was hosed because I was selfish with my intellect. the world's biggest self-pwn serves me right that I didn't get a job in the industry, and missed out on all that dough, doesn't it? the direction hasn't changed. Its going the same way and you should think twice about subscribing especially when once you ask anything, - i assume - the provider is obliged to scrub the request for safety concerns, and log it and keep an archive, and casually absorb it.... and in the mean time - if I understand the nature of the world correctly - anything you ask/submit actually no longer belongs to you. [q.v. FBook T&Cs] there is no privacy in your queries and there will be no guarantee that what you may have ASSUMED would be a unique innovative idea all your own is now for all in the hands of a much smarter thing that has a lot better access to resources and countervailing data it may decide not to tell you about. this doom has been ongoing for a long time. its bad, and we should start to figure the least bad outcomes and work toward those first |
|