brainwad

3 ms·

1.
by brainwad 9d ago
There are a lot of SKUs to regulate. The cost of the bureaucrats checking the requests for basic "my price is going up because my inputs went up" price rises would be staggering. Also it would disincentives any price drops being p
2.
by brainwad 10d ago
Yeah, but on the other hand the borrowing is relatively small compared to the company, so even if the company shrinks by 90% it will still be serviceable. Many western governments including the US have ridiculous debt loads at multiples of
3.
by brainwad 11d ago
Italy recently issued USD-denominated bonds, which are directly comparable, and the yields are much higher. The 30y ones are at 6.21% YTM now, vs 5.37% for US 30y Ts. For comparison, Alphabet borrows cheaper than Italy: 6.02% on 2060 maturi
4.
by brainwad 11d ago
I mean, sure, assassins and terrorists are also responsible for their actions. But we still try to prevent them structurally.
5.
by brainwad 11d ago
That's not how abliteration works. The trainer invariably put effort into making the model not dangerous, precisely because they want to be accountable. But because they release its weights, someone can come later and do "weight s
6.
by brainwad 11d ago
No it won't, because abliterated open weights models exist and unless you try to censor the internet they can't really be withdrawn after publishing. This is exactly the problem that the labs are proposing to fix: a dangerous mode
7.
by brainwad 11d ago
If your security model is having to imagine all the ways your frontier models might misbehave in novel ways and preemptively sandbox them, you don't have a security model. The only way that will work is general alignment.
8.
by brainwad 11d ago
The problem is not per se that it was Hugging Face. It's the wild overstepping of reasonable bounds by itself without any human consultation.
9.
by brainwad 11d ago
The hugging face hack shows otherwise, no? Unless you think humans are at fault even for secretive, autonomous, non-prompted behaviours of their AIs, in which case it's just semantics.
10.
by brainwad 12d ago
It's not illegal to write a similar book or song to an existing one. Copyright only protects existing works from literal copying (possibly in part).
11.
by brainwad 12d ago
Well, adblockers don't copy the copyrighted content. They just control how it's rendered on the user's machine. Copyright cares about making copies and especially distributing them.
12.
by brainwad 12d ago
They can just round up some friendly users and sue under their names. Starting with their own corporate accounts?
13.
by brainwad 12d ago
It's depriving X of impressions that they could monetise, no? Xcancel doesn't have to make money itself, it just has to impair the rights of the copyright holder. Otherwise piracy would also be legal as long as it were non-profit.
14.
by brainwad 12d ago
That's exactly the way UK courts are heading, see Getty vs Stability AI. The court ruled that there's no infringment because the model doesn't store exact copies, just derived weights, and therefore when it generates new imag
15.
by brainwad 12d ago
There's a difference between creating a market for something better, so that nobody wants the old thing, and competing _in_ the market for the old thing by copying it directly.
16.
by brainwad 12d ago
I think that's more true in image generation than in text? At least, all the money is in LLMs that write code, not LLMs that write O'Reilley-style coding books.
17.
by brainwad 12d ago
Bartz v Anthropic. Though the plaintiffs did get something, it was because of the piracy to the original works (competing against the legal market for the books), not the use of them to train the LLM.
18.
by brainwad 12d ago
Precedent is pretty clear: competitive uses bad, transformative uses good. Xcancel scrapes and then competes directly with X, whereas LLM labs scrape the internet to make an agentic intelligent bot, a transformative use of the scraped conte
19.
by brainwad 13d ago
Human growth has already been exponential, so it seems absurd to assume it would be only polynomial.
20.
by brainwad 13d ago
Musk being in the conversation is a great justification for Anthropic accelerationism. Who wants to live in the world where _Musk_ controls the AGI?
21.
by brainwad 13d ago
It's not misinformation. By default Google has been able to click track with JS. If you disable JS, they don't serve you search ( https://serpapi.com/blog/google-now-requires-javascript/ ). Yes, if you ar
22.
by brainwad 14d ago
Unfortunately copyright law allows them to give something to one person without giving it to others _and_ to sue to the former person to block them from giving it to the others.
23.
by brainwad 14d ago
Well, that's rather the point, right? Kagi and other micro search engines sell a product that just resells Google search results while telling people it's a premium product better than Google. Obviously Google isn't pleased a
24.
by brainwad 14d ago
Didn't Google take it down because of copyright? With the rise of paywalls it became a backdoor around them, but Google doesn't actually have a license to republish the copyrighted text they scraped.
25.
by brainwad 14d ago
Adding -ai to the query works fine. Or like the other responder said, use the web tab instead of the all tab.
26.
by brainwad 14d ago
That flag is the result of clicking on the "Web" tab of the results page, so it kinda is a documented feature. All the tabs have their own udm value (e.g. image search is 2, AI mode is 50, etc.).
27.
by brainwad 14d ago
Because Google search isn't a hypertext document, it's a web app that presents dynamic ephemeral content to the user. There's no need for stable hyperlinks, since either you will click them within a few seconds, or they will
28.
by brainwad 14d ago
It seems feasible for state actors, at least: https://en.wikipedia.org/wiki/GNSS_spoofing#Ocurrences
29.
by brainwad 14d ago
He wasn't the one who merged ads and search into the same org. The blame has to fall on Sundar, surely. > promoted to a cushy do-nothing role The people who this happens to are not perceived as successes. Everyone knows they are gen
30.
by brainwad 14d ago
They own the JS on the page, they don't need the redirects to know where you clicked...

More ›