cs702

186 ms·

2431.
by cs702 6y ago
Wow, there's so much... negativity in this post by Rogozin. It reads like "sour grapes": " Driven by hunger, a fox tried to reach some grapes hanging high on the vine but was unable to, although it leaped with all its
2432.
Coronavirus hospitalizations rise sharply in several states(washingtonpost.com)
7 points by cs702 6y ago|0 comments
2433.
by cs702 6y ago
We entered this crisis with record levels of household, corporate, and government debt worldwide. Quoting from a World Bank report warning about record global debt levels on December 19, 2019 -- three months before the lock-downs: > The
2434.
How Things could go wrong in America(ft.com)
2 points by cs702 6y ago|1 comments
2435.
by cs702 6y ago
Have you looked at capsule-routing algorithms that use expectation-maximization to generate layer outputs in deep neural networks? For example, in https://research.google/pubs/pub46653/ and https://arxi
2436.
by cs702 6y ago
This reads like something that normally would be written about an underdeveloped country, which, lacking established institutions and rule of law, is run by a kleptocracy for the kleptocracy. Yet we're talking about the United States.
2437.
by cs702 6y ago
There are a lot of people from Microsoft on HN. Hopefully at least one of them will respond here, or even better, look into righting this wrong.
2438.
by cs702 6y ago
Yes, mind-boggling. For each task, the authors feed a context window of text with either zero to a few sample queries and responses, followed by a query without the response. The model generates a response for the last query. BTW, thi
2439.
by cs702 6y ago
> No the zero-cost refers to the abstraction (and runtime cost), which still is zero-cost. Deallocating is part of the normal work load not the abstraction. Yeah, you're right. In hindsight my comment was poorly thought-out and poor
2440.
by cs702 6y ago
Yeah, you're right. In hindsight this was a poorly thought-out and poorly written post on my part.
2441.
by cs702 6y ago
The idea of "marketing Haskell" to me is akin to marketing math and mathematical notation. Yes, Haskell is elegant. Yes, it is powerful. Yes, it is succinct. Yes, the benefits of its algebraic type system are many and very real --
2442.
by cs702 6y ago
In other words, Rust's automagical memory deallocation is NOT a zero-cost abstraction: fn get_len1(things: HeavyThings) -> usize { things.len() } fn get_len2(things: HeavyThings) -> usize { let len = things.l
2443.
by cs702 6y ago
> "Why will this paper matter in 5 years?" 3. First evidence that performance continues to improve at hundreds of billions of weights -- paving the way for trillions of weights, approaching orders of magnitude comparable to tha
2444.
by cs702 6y ago
I disagree. If as a researcher you come up with a new kind of model that (1) performs better than small-scale transformers given the same computational budget , and (2) improves its performance as you increase the computation budget (mode
2445.
by cs702 6y ago
I never said otherwise. Like physicists, we need both new theory (for new insights, ideas, models, etc.) and new experiments (for replication, performance, scalability, etc.) Also, Moore's Law is running on fumes, yes, but there's
2446.
by cs702 6y ago
GPT-3 was pretrained on five datasets (Common Crawl, WebText2, Books1, Books2, and Wikipedia; see table 2.2), and then used on previously unseen tasks (Q&A, translation, cloze, etc.) without finetuning , i.e., weights were not update
2447.
by cs702 6y ago
All valid points, but I disagree with the conclusion, for several reasons: * First of all, the GPT-3 authors successfully trained a model with 175 billion parameters. I mean, 175 billion. The previous largest model in the literature, Google
2448.
by cs702 6y ago
Thanks. I should have written "if using Float32," which is what I meant -- instead of "with Float32," which in hindsight reads a bit ambiguous. But regardless of which floating-point representation is used, the number of
2449.
The Bitter Lesson (2019, relevant in light of GPT-3)(incompleteideas.net)
1 points by cs702 6y ago|0 comments
2450.
by cs702 6y ago
This looks like a big deal to me: 1. First of all, the authors successfully trained a model with 173 BILLION PARAMETERS. The previous largest model in the literature, Google’s T5, had "only" 11 billion. With Float32 representation
2451.
by cs702 6y ago
Good question. The approach I proposed works for production applications that require an easy-to-replicate, deterministic environment, but I wouldn't recommend it if you're trying to build, say, Python packages or frameworks meant
2452.
by cs702 6y ago
It's evident from reading the OP and previous similar posts on HN that many developers find it difficult to specify and replicate deterministic Python environments for their applications. Personally, I have found it best to use (a) a v
2453.
by cs702 6y ago
No matter how you look at it, Boeing's business appears to be in terrible shape. The Atlantic has a decent article on how the company lost its way: "a company once driven by engineers became driven by finance."[a] However, fr
2454.
by cs702 6y ago
The OP believes the covid-19 crisis has shown us "many truths we've long ignored," and that once the pandemic is over, humankind will come together to address them. Among the long-ignored "truths" mentioned by the O
2455.
Watching the coronavirus-19 crisis unfold from a Turkish prison, I’m hopeful(washingtonpost.com)
2 points by cs702 6y ago|1 comments
2456.
by cs702 6y ago
First, health crises spreading everywhere. Then, almost simultaneously, economic crises spreading everywhere... but luckily for everyone, governments around the planet step in to keep it from getting worse. Then, an incipient financial shoc
2457.
by cs702 6y ago
Keep in mind, a ton of people and companies are asking for this, because they genuinely want to use hardware-accelerated PyTorch and TensorFlow "through a Linux-thingy" on Windows. BUT I MUST AGREE WITH YOU that, over time, this
2458.
by cs702 6y ago
Likely to be followed by The Washington Post, the Wall Street Journal, Bloomberg, and other leading news organizations in the US, as well as elsewhere. High-quality news and editorial content is sought out mainly by smart, educated, and&#x2
2459.
by cs702 6y ago
> Project Reunion is our vision for unifying and evolving the Windows developer platform to make it easier to build great apps that work across all the Windows 10 versions and devices* people use.* I'm always a bit skeptical of pr
2460.
by cs702 6y ago
Wow, the entire advertising chain here looks like a completely unnecessary TAX on restaurants and consumers: * A restaurant pays the advertising fee to DoorDash/GrubHub but the money is used to divert organic searches for the restaura

More ›