anankaie

8 ms·

31.
by anankaie 2y ago
Moreover, you get surprisingly out-of-class (size-wise) performance if you fine-tune for your specific problem space. Even if you only train in a parameter-efficient way.
32.
by anankaie 2y ago
Hear me out here: Maybe if we split the address space into various use-specific segments...
33.
by anankaie 2y ago
No, the internet is made of daemons
34.
by anankaie 2y ago
Archive link: https://archive.today/WyFBG
35.
by anankaie 2y ago
One can only hope, though given how much more authoritarian everyone (in the US, at least) has gotten, I am not holding out much hope. Good luck fighting this; I mean that sincerely.
36.
by anankaie 2y ago
Motte-and-Bailey speaks to equivocating about positions in argument. There are many other ways to equivocate, e.g. over the meaning of a single word, used in multiple ways in the same argument, as opposed to a retreat to an easier-to-defen
37.
by anankaie 3y ago
Yes, but the original author is incentivized to attempt to show where the replicators got it wrong, so there will still be a push to correct the bad data from the false replication failure. With that said, I am convinced it will be a panace
38.
by anankaie 3y ago
You're right. We should make it so they have to pay for the energy they consume. Maybe price it by kilowatt-hour or something.
39.
by anankaie 3y ago
And most ML models cannot do that either. This is not a good analogy.
40.
by anankaie 3y ago
https://archive.is/NCzks
41.
by anankaie 3y ago
The problem is that there is no way to trust that Unity will not alter the deal in an underhanded way in the future. I do not want my purchases beholden to their capricious whims. I sympathize with the devs, but sometimes it turns out that
42.
by anankaie 3y ago
The problem is that the "infinitely small" is already defined, and not equal to zero: The infinitesimal. It is not so simple as defining what the result should be - one needs to make sure this is consistent with the rest of mathem
43.
by anankaie 3y ago
Apple does not have a search engine competing with DuckDuckGo, though, unlike Google. The whole point of it was to avoid antitrust rulings for referencing their own search engine. If they force other search providers to pay for placement, t
44.
by anankaie 3y ago
> Previous testimony in this case is pretty damning for Google, since it implies their ad business is built on multiple levels of fraud. I have not been following this as closely as I should have. Is there a specific source of that testi
45.
by anankaie 3y ago
Satire and commentary is covered under Fair Use.
46.
by anankaie 3y ago
I find that entire Usenet conversation very interesting for all sorts of quotes, far and beyond what is mentioned in the blog post about Perl/RE: > [...] In particular, I was baffled that > there was no mention _whatsoever_ of ad
47.
by anankaie 3y ago
While I agree with you on the relation of GP's Hyp1 and Hyp2, you are making an unfounded assumption of a sampling process being necessary to perform human speech. I do not believe we have the understanding of how thought is represente
48.
by anankaie 3y ago
> “The problem isn’t Kathleen Kennedy.” I am not entirely sure that matters. Pissed-off fans undoubtedly want to see metaphorical blood for what, to their eyes, was a forced, ideologically-driven tarnishing of their childhood memories. A
49.
by anankaie 3y ago
It sounds like Cowen knows exactly what needs to happen to reduce the malaise of "quiet quitting" - show workers that their earnings can/will grow to account for inflation and whatever productivity increases they show, and si
50.
by anankaie 3y ago
I would go further: noreferrer should be the default. Alternatively, let each user configure it, as a privacy option.
51.
by anankaie 3y ago
Maybe a better way to generalize would be "when not at the top of the world"
52.
by anankaie 3y ago
Yes, and more broadly $Company is at their best when they are struggling.
53.
by anankaie 3y ago
And then there is cork
54.
by anankaie 3y ago
Not quite. The underlying technology is open-source, but the specifics of the models of late (to say nothing of the tuned parameters) are not. Just because the core of the architecture - or even the tooling used to build it - is open source
55.
by anankaie 3y ago
And in large part because FDA multi-stage trials are very expensive to get through and the risk of failure is high. I know it is popular to hate on the pharmacy companies right now - and deservedly so - but the research costs are not totall
56.
by anankaie 3y ago
The issue is that the admins only selectively enforce their rules, and are very ideologically driven at that (see the general rule of not allowing bans for other subreddit participation). This sets the tone for the moderation.
57.
by anankaie 3y ago
If it was truly random, it would not generate useful outputs. You can think of it similar to randomly sampling one of the authors' (of the training data) observed ideological lean. One way to model and measure this would be to think of
58.
by anankaie 3y ago
Alternatively, could we consider treating the technology as a potentially dangerous technology without giving politicos all over the world additional ammunition and incorrect mental models with regards to the regulation thereof? To be clear
59.
by anankaie 3y ago
> its own agenda Minor nit: ChatGPT has no agenda beyond a statistical sampling over its training set. The developers at OpenAI, on the other hand, do.
60.
by anankaie 3y ago
It is really weird to me that OpenAI, of all companies, does not understand that the internet is forever: https://archive.is/rcbem

More ›