Majromax

24 ms·

271.
by Majromax 2y ago
The math seems off to me here. In particular: >> So here is the deal: If you have 16,000 GPUs, and you have a blended average of on-demand (50 percent), one year (30 percent), and three year (20 percent) instances, then that works ou
272.
by Majromax 2y ago
First, their trading's global, so "for each average American" isn't the right denominator, nor is it obvious to divide things on a per capita basis rather than a wealth-weighted one. Second, it's not at all obvious
273.
by Majromax 2y ago
Current above-the-fold headlines on https://www.theonion.com/politics : * Biden Carried Away By Ants * Janet Yellen Unveils Plan To Boost Economy By Stealing World’s Largest Diamond * Congress Quickly Passes Funding For Nati
274.
by Majromax 2y ago
> How do I monitor whether the contacts are closed or open? I assume with a GPIO pin, but I’ve never been able to google this question and find anything of use. Without making assumptions about the microcontroller used, attach ground to
275.
by Majromax 3y ago
Also look at the prime-age participation rates ( https://fred.stlouisfed.org/graph/?g=1k7Vy ), which help control for demographic changes over the decades. The 1963 prime-age LFPRs were 44/99% for women/men re
276.
by Majromax 3y ago
> That's missing forest for the trees. The more important part is that average 3.7 people household was a single-income one. That's not necessarily true. Per FRED ( https://fred.stlouisfed.org/series/LRAC25
277.
by Majromax 3y ago
Matrix-scan keyboards allow for an unlimited number of overlapping keystrokes, but that allowance requires per-key diodes to prevent ghosting. Inexpensive keyboards often forego the per-key diodes in favour of cheaper/thinner constru
278.
by Majromax 3y ago
Proper betting shops are market makers, ideally balancing their books such that they profit by the house spread regardless of outcomes. A bookie that takes on a substantial net position is accidentally gambling with their own money.
279.
by Majromax 3y ago
> And they don't add any fees, you just pay the duties and taxes. Are you sure about this? Canada Post's webpage ( https://www.canadapost-postescanada.ca/cpc/en/support/articl... ) says: >>
280.
by Majromax 3y ago
It's likely a result of the interplay between the image generation and caption/description generation aspects of the model. The earliest diffusion-based image generators used a 'bag of words' model for the caption (see
281.
by Majromax 3y ago
To summarize the article: using backpropagation to train a small LSTM to accept strings of the form (a^n b^n), regularization with both L1 and L2 losses result in loss spaces where the loss-optimal solution is not the generally-correct one.
282.
by Majromax 3y ago
> That seems like a problem in our legal system though. The civil legal system is designed to settle essentially private disputes between parties. It is not really a matter of public interest whether my neighbour or I bear the liability
283.
by Majromax 3y ago
> The author seems to be mistaking the benefit of selection effects with competence. [...] The big platforms actually are radically more competent at anti-fraud / anti-spam / etc, but the scope of their operations make them su
284.
by Majromax 3y ago
> What am I missing? I'm a bit confused about the accounting details as well, but I think the key bit is that a scrapped movie becomes an immediate tax write-off, whereas a released move must depreciate its costs over time. By wri
285.
by Majromax 3y ago
> everybody only wants to spend $0.50 and then they complain when they get $0.50 worth of widget. I don't think that's quite the case. Instead, I think the people complaining at receiving the $0.50 widget complain when the (n
286.
by Majromax 3y ago
> I'm curious: what is the rationale for "in an incognito tab" being part of the test harness? It's the closest we can easily get to the 'average user experience'. Someone who has a long account/cookie
287.
by Majromax 3y ago
> I find the results of that famous case completely insane. A "thumbs up" emoji doesn't always mean [...] I invite you to read the decision itself ( https://www.canlii.org/en/sk/skkb/doc/
288.
by Majromax 3y ago
There is some pressure in this direction for respiratory illnesses. It's easiest to spread an illness when it's present in the upper airways (and thus easiest to cough/sneeze out), but the illness does more damage to the lun
289.
by Majromax 3y ago
That's discussed at the end of page 4, and it comes from the Noci et al (2023) reference.
290.
by Majromax 3y ago
> Let's assume about 1 word per token and 5 characters per word. Furthermore, US-ASCII tokens requires one byte per character in UTF-8. So, that gives about 40 petabyte. Thanks, I was looking for that estimate but could not quickly
291.
by Majromax 3y ago
> I bet if you add up that data it's a lot. Let's Fermi estimate that. A 4k video stream is about 50 megabits/second. Let's say that humans have the equivalent of two of those going during waking hours, one for visio
292.
by Majromax 3y ago
> You tell me crashing was expected default behaviour to a point where the suitability of implementation details was decided by how much one would crash? No, and Chen emphasizes this point repeatedly in his blog post. The customer knew
293.
by Majromax 3y ago
Sure they can. DNS black holes can only work when ads are served from a known and distinct set of domains than than the intended website content. If the domains are the same, then one cannot distinguish between an ad and content at the DN
294.
by Majromax 3y ago
I wonder if this is some cross-pollination legacy of Windows Phone. On such a device, background processes are still necessary, memory is a much tighter constraint, and enough pages might be mapped directly from device flash for auto-evict
295.
by Majromax 3y ago
> lets Apple know what the defect rate is, which for some reason the manufacturers are reluctant in telling Apple My suspicion is that an unknown but nontrivial share of the 'defects' were of the 'fell off a truck' va
296.
by Majromax 3y ago
> Could it be that the fine tuning is causing these? The authors suggest that the fine tuning isn't the problem for a couple of reasons: * First, if it were just the question phrasing, then we'd expect the model to do symmetric
297.
by Majromax 3y ago
Per the paper, each of the tested facts was included in the fine-tuning with 30 paraphrased examples that preserve the testing order. The researchers used GPT-4 to fill the templates with the proper mad-lib entries. The first few "des
298.
by Majromax 3y ago
> At first glance this doesn't seem that surprising. We often use "is" in a way which isn't reversible. e.g. They appear to only be testing the 'reliable' cases. There schematic example was fine-tuning the
299.
by Majromax 3y ago
What you've done here is tell SymPy to use extra precision for the intermediate (and final) output. This doesn't truly fix the problem of cancellation and loss of precision, but for many practical purposes it can postpone the p
300.
by Majromax 3y ago
Floating point values are binary scientific notation with a fixed precision. All of the unintuitive behaviour of floats is repeated if you imagine conducting decimal calculations yourself in said scientific notation, with intermediate roun

More ›