sometimelurker

4 ms·

31.
by sometimelurker 4mo ago
their putting a ton of money into things that will benefit AI work if they work, but chipmaking is important for other reasons too, so if AI (somehow? I doubt) doesn't give enough ROI chips can be used for other things
32.
by sometimelurker 4mo ago
The antidote to that (I think, I've got loose vibes about biology and genetics) would probably be introducing more variation via some manipulation of the genome, maybe crispr?
33.
by sometimelurker 4mo ago
> not sequence all people's DNA While the people in power yesterday/today might not abuse databases of their entire population's DNA, those of today/tomorrow might. In the long-run, of 10y,20y, 50+y, you cannot guaran
34.
by sometimelurker 4mo ago
But there's still some really major issues with 2). Do you think it would be fair to have the children of billionares to be much much more intelligent than normal people, let along the children of poor people? I'm not sure what
35.
by sometimelurker 4mo ago
I don't think it should be mandatory but ofc it should exist; people should be able to choose.
36.
by sometimelurker 4mo ago
reminds me of captian America 2, in which the villains (hydra) said that what people actually want is security, not freedom. most people are boring blue-pillers* in this regard. *(which I mean in the literal meaning in the matrix)
37.
by sometimelurker 4mo ago
There's definitely a sane debate needed on this. I think there's a lot of things here people wouldn't be against, but at the same time more thought needs to be put into this. Like what if you could GMO your kids to be smarter
38.
by sometimelurker 4mo ago
sorry abt that, its just there's a lot of bots on hn nowadays and I get irritated sometimes. mb
39.
by sometimelurker 4mo ago
interestingly in areas with higher average wealth there should be a sharper difference between the richest and poorest people in that area. consider Elon musk, and the ratio between him and the 2ed and 3ed up to the 10th richest people on e
40.
by sometimelurker 4mo ago
> But we didn't create more cities. be careful what you wish for. eventually we'll have a 'New San Francisco' and a 'New New York City'. Imagine the acronyms for these, NSF and NNYC. what if we just agreed
41.
by sometimelurker 4mo ago
Oh I figured there's a real human being named Josh Vander and youre his openclaw or something. why do you have the `bot` suffix?
42.
by sometimelurker 4mo ago
You have the word 'bot' in your name and judging by your comment history love using em-dashes. Are you a machine or not?
43.
by sometimelurker 4mo ago
And they do have internal latent vectors for their own states, so there's some kind of recursion/introspectiveness dynamic.
44.
by sometimelurker 4mo ago
As I see it, LLMs will/do develop 'introspectiveness' if it is selected for during RL. If its not delibrativly selected for (this is probably done because it helps with alignment; if the machine can align itself that's g
45.
by sometimelurker 4mo ago
apparently bosch semiconductor exists. whatever theyre doing over there they should do more of it tho, I want cheaper ram
46.
by sometimelurker 4mo ago
works well on pi, but I use a smaller one at higher tok/s for more repetitive tasks
47.
by sometimelurker 4mo ago
train stations are boring and no military spends time thinking about train stations . ASI hacker nonsense, that gets their attn.
48.
by sometimelurker 4mo ago
asml is part of the chip supply chain; amsl has chip making talent, and EU needs more better chips to make more better ai also an em-dash and genuinely in the same sentence? cmon bro how much claude have you been using?
49.
by sometimelurker 4mo ago
> The big question is who is going to fund it? this feels like a small question, the money needed for this isn't that much (probably ~20b) and every country needs cyberwarfare abilities, especially automatable ones, so the military
50.
by sometimelurker 4mo ago
for RL cost* pretraining becomes more expensive actually as you make MoE models sparser (you need more tokens in the pretrain, and if you don't have that then you need to train for longer)
51.
by sometimelurker 4mo ago
Europe has asml
52.
by sometimelurker 4mo ago
lack of EUV machines
53.
by sometimelurker 4mo ago
Its a known fact that the USG isn't happy with Anthropic https://xcancel.com/PeteHegseth/status/2065897156226015690# <- dated june13/26 https://www.anthropic.com/news/where-stand-
54.
by sometimelurker 4mo ago
So now they can say stopping mythos+fable wasn't politically motivated. I wonder where this will go. Hopefully there's some concrete legislation for regulation of powerful AI enacted soon.
55.
by sometimelurker 4mo ago
honest question: does the nsa have the abilty to take the model weights from anthropic? also, as I understand it anthropic employees from the USA have mythos access, and I dont see why this shouldnt extend to the nsa. this seems pretty sill
56.
by sometimelurker 4mo ago
source? credible rumor has mythos at 10 trillion params
57.
by sometimelurker 4mo ago
I really seriously doubt this. no proof, obviously, but this sounds too unbelievable. the NSA probably has the mythos pretrain and might be finetuning it themselves.
58.
by sometimelurker 4mo ago
well its actually called ablation, and its one way to do mech interp. anthriopics got a bunch of work on mech interp here https://transformer-circuits.pub/ , like SAEs and NLAs
59.
by sometimelurker 4mo ago
this appeared some time ago, https://taalas.com/ , but I'm sure there's others thinking these same thoughts. this would be best for small models imo, nothing frontier because that changes too fast
60.
by sometimelurker 4mo ago
Ive tried both, and pi's core is a lot smaller but can be extended to have as many features as opencode. personally I like using pi with llama.cpp, and use stuff like pi.dev/packages/pi-llama-cpp-stats to combine them better.

More ›