marcinzm

46 ms·

691.
by marcinzm 3y ago
Amazon has had very strong competition that they mostly beat. Meta has faced strong competition which they beat through scaling out acquisitions. Apple has gone from nothing to the world's richest company. Microsoft was a joke and now
692.
by marcinzm 3y ago
In my experience the key difference is that managers often accept they are part of an inherently soulless capitalistic exchange of their time for money. Many ICs prefer the illusion that they're not. Everything else tends to flow from
693.
by marcinzm 3y ago
Give me an example of a large (10k+) company that uses a better approach and has better business metrics than FAANG. Otherwise you're talking hypotheticals while FAANG has continuing growth and 20+% profit margins.
694.
by marcinzm 3y ago
I find it amusing that someone making $500k+ at Meta would consider themselves a pleb.
695.
by marcinzm 3y ago
The context of this thread is the cost of training brains and models on comparable tasks. Not that the model is comparable to a human in every way. If you want to be pedantic then 6% of the human brain is the visual cortex but then you also
696.
by marcinzm 3y ago
Sure but my point is that the energy costs are in the same magnitude. If you want to be pedantic then only 6% of the human brain is the visual cortex. But AlexNet is also an inefficient model so something like an optimized ResNet is 100x as
697.
by marcinzm 3y ago
It takes basically a week on a single GPU to train AlexNet which has human level ImageNet performance. Let's say it's 500 W for the GPU versus around 10 W for a human brain. So that's 84kwh for the model and 175kwh for the ba
698.
by marcinzm 3y ago
The whole argument is that at large scales with billions of params it doesn’t matter specifically because of those billions so giving a toy example seems to miss the point.
699.
by marcinzm 3y ago
The code of the paper is all on Github so you can verify that line of thinking if you want to.
700.
by marcinzm 3y ago
>The latter depends very strongly on how much computation is needed to compute those embedding vectors. Sure but the gzip metrics are worse than FastText which computes the embeddings in essentially no time. Tokenize, lookup embeddings b
701.
by marcinzm 3y ago
But FastText (2015) already exists and beats this gzip approach on all criteria. So the invitation has already existed before BERT (2018) and continues to exist.
702.
by marcinzm 3y ago
Sure but it's existence means the statement is really "gzip approach not better than dnn models, and doesn't compete or be cheaper to run than previous models like FastText." That's not a very meaningful value state
703.
by marcinzm 3y ago
>gzip approach not better than dnn models but mostly competes and much cheaper to run Does it? It looks to do worse than FastText in all benchmarks and kNN is not a cheap algorithm to run so it might actually be slower than FastText. edi
704.
by marcinzm 3y ago
Looking at charts of the US gold supply it seems to have grown fairly significantly in the early 1900s up to the great depression, and seems to have been more than the 60% cited for money supply growth.
705.
by marcinzm 3y ago
Looking at charts the money supply mainly grew during WW1 and growth went back to the previous levels after WW1.
706.
by marcinzm 3y ago
I "love" meetings where they just recite word for word cells from an Excel spreadsheet for an hour.
707.
by marcinzm 3y ago
>I suppose we should ask visual impired person instead of guessing. You mean like the one who presumably wrote the linked article that non-blind people are arguing against the suggestion from?
708.
by marcinzm 3y ago
>As I wrote elsewhere in the thread, if EU didn't word it the way they did, most OSS projects would simply slap "not fit for commercial purposes" in the README next day after CRA became law with a smug smile. So? Seems an
709.
by marcinzm 3y ago
>Your project needs to make such an impression. And how much are the legal fees going to be if you are accused of this and need to defend yourself? Given the fees if found guilty this seems very much a safer to just not risk it type of s
710.
by marcinzm 3y ago
So if a company decides to open source some of their internal software and periodically release updates they'd be liable now? I don't see how that would not outright kill 99.9% of the open source released by companies. Most compan
711.
by marcinzm 3y ago
So it seems to be a projection based on a presentation by the CEO of Microsoft that said the loss of potential revenue and growth is $3T in 2016 (note this is itself not actual measured impact but also a counter-factual projection). In othe
712.
by marcinzm 3y ago
>The problem is that whether the project is hobby or not is judged from the side of the consumer, i.e. if the software is usable "in the course of a commercial activity" (for the user). Have you seen the dependency trees for co
713.
by marcinzm 3y ago
That's more than the worldwide spending on IT per year.
714.
by marcinzm 3y ago
>OSS amount to 5.5 trillion EUR every year So the EU loses over 30% of it's total GDP to security vulnerabilities? I somehow find that figure very suspect and using clearly exaggerated numbers as justification makes me distrust the
715.
by marcinzm 3y ago
That's true but in my experience it's more often that the management chain tries to implicitly not have the ICs care about the product or customer. They may not even realize consciously that they're doing twenty little things
716.
by marcinzm 3y ago
A lot of the essay is along the same lines to me. Top down product strategy that teams then simply execute on versus being co-creators of. In that environment you want rituals/routine since it avoids the team thinking about things that
717.
by marcinzm 3y ago
Other have noted how bureaucratic this seems to be for a 200 person company but to me it also seems so top down/hierarchical. Strategy that flows top down with little visibility, input or involvement from ICs except when forced to avoi
718.
by marcinzm 3y ago
If A then B does not mean if B then A.
719.
by marcinzm 3y ago
If your goal is to make a lot of money (not just some money) then using VC money to get there is a very reasonable play. Just siphoning some money from later rounds could give you tens of millions. You don't even need to succeed after
720.
by marcinzm 3y ago
Google building a social media product. That will definitely go well this time around.

More ›