blagie

47 ms·

541.
by blagie 3y ago
I have plenty of pieces of code I've written, decades ago, where I know what they do, but have NO idea how they work. If I want to understand a piece of code, I need to read it and understand it. Modern models have billions of paramete
542.
by blagie 3y ago
> We know exactly how these algorithms work. We have no idea how these algorithms work. > We know every calculation - the maths is not particularly difficult, We do know that. > we understand how the training process leads to infor
543.
by blagie 3y ago
A lot of previous plateaus in AI are usable and used by billions daily, for example, giving good navigation routes on your phone, managing NPCs in a video game, showing ads, or recommending movies. It's not that they don't have va
544.
by blagie 3y ago
I think it's a mistake to assign human emotions and behaviors to robots. Human emotions and behaviors were, depending on which side you fall on, a result of: - Millions of years of evolution to have as many babies as possible and to ha
545.
by blagie 3y ago
I think if things went that far, OpenAIs valuation would very quickly pop to zero. Real MSFT stock beats a theoretical could-have-been with a 10x upside.
546.
by blagie 3y ago
I don't like the cynicism here. * Yes, performance numbers would be helpful. * No, this is not new, and for most applications, not practical. The basic point is that yesterday, I couldn't run a large model locally. Today, I can, i
547.
by blagie 3y ago
With very rare exceptions (e.g. nuclear weapons), taxation is almost always a better mechanism than outright bans. Even for pretty bad things (e.g. really toxic pollution), the taxes just need to be set obnoxiously high. By "obnoxiousl
548.
by blagie 3y ago
> But I do think there might be an economic argument in favour of at least some advertising to bring awareness of new market entrants, and especially of new categories of product/service. That's the standard argument made: Cons
549.
by blagie 3y ago
The unspoken secret here is that ads distort markets, suck up resources, and thereby make economies less efficient. Banning them would be an economic boon. There. I said the quiet part outloud.
550.
by blagie 3y ago
I suspect there are a few issues: 1) NVidia drivers had a lot of secret sauce to give high performance. 2) NVidia for machine learning still has a lot of platform lock-in (although fading gradually), and cross-compatibility doesn't hel
551.
by blagie 3y ago
I will ignore the potentially false statements and restrictions on arm sales, and answer the high-level question: Force design is a question of what the force will be used for. A military for a counterinsurgency in Iraq, will look very diff
552.
by blagie 3y ago
Educated opinion: Everything I've seen around the F35 seems to indicate it overshot expectations and is a very good plane. Uneducated opinion: Personally, I think it's also overused to a silly extent, but that's a question th
553.
by blagie 3y ago
Nukes are not within anyone's reach, or instead of teen suicides and school shootings, we'd have nuclear craters in place of some cities. Most of these people: https://www.forbes.com/billionaires/ Could not a
554.
by blagie 3y ago
Let me give a short list of technologies I wouldn't want accessible to everybody: - Thermonuclear warheads - Designer biological viruses and bacteria - Tools to remotely compromise / brick all computers in the world ... and so on.
555.
by blagie 3y ago
I drive the cheapest possible car. I'd seriously consider paying double that for one which was genuinely open-hardware. I'd never consider Tesla since the way it works, you buy the physical hardware, but in practice, Tesla control
556.
by blagie 3y ago
Step 1: Buy a VPN subscription. It's about $3/month Step 2: Run script I'm a lot happier to pay for VPN and hard disks than for Google's bait-and-switch games. Youtube and Google became popular based on a social contract
557.
by blagie 3y ago
The timing is right. With 2/3 of the company grumpy, it'd be untraceable.
558.
by blagie 3y ago
Let me clarify: This wasn't intended to contradict the root-n limit. Both limits exist. * The root-n limit is based on physics of a 3d universe. * A log n limit is based on computation, and would apply even in a infinite-dimensional un
559.
by blagie 3y ago
FYI: This article sucks (in relative terms) if you read Part I in isolation. Click on Part II to get to the really good parts. From there on, it gets weaker. If you're busy, you can skip / skim parts III and IV. My hope is this po
560.
by blagie 3y ago
FYI: I ran my text through GPT before I posted, which is why I could tell you which models read this correctly, and which incorrectly. You may also wish to look at other responses to my post.
561.
by blagie 3y ago
Thank you. I appreciate that you responded. Your free reading assessment: Your reading comprehension is similar to smaller and first-generation GPT-3 models. It is below that of more recent GPT-3 divinci models (e.g. text-divinci-3), gpt-3.
562.
by blagie 3y ago
Take the comment above as an example of my work. Read it again. What was the writer disappointed by? (That's not a rhetorical question. That's a reading comprehension question. Please post your answer)
563.
by blagie 3y ago
Well I, for one, was completely disappointed. I was just being drawn into the plot. Technically, it was the best Blender animation I'd seen.... ever. The foreshadowing was perfect. I knew something awesome was coming.... And boom. It s
564.
by blagie 3y ago
> I don't know many people who would claim that with a straight face. But it is your right to have that opinion. I think you're taking the opinion out-of-context. The question was: "What are good places to get cheap colleg
565.
by blagie 3y ago
I'll make this very concrete. I mentioned MIT. The largest community college in Massachusetts, Bunker Hill, has the following math courses: • MAT-093 Foundations of Mathematics • MAT-097 Foundations of Algebra • MAT-09
566.
by blagie 3y ago
Most, but not all, do. However, not all BS degrees were created equal. Seriously. Look at the requirements of /any/ community college BS degree, and compare to the same degree at /any/ elite university. Please. Post a li
567.
by blagie 3y ago
It's a fine strategy, but the critical words there are "transferring to the state university later." Critically, a community college will NOT cover beyond the first 2-4 semesters of a university BS degree. This is also not a
568.
by blagie 3y ago
Super-spoiler: This is absolutely, completely incorrect. 1) Teaching quality doesn't correlate much with quality (if anything, the inverse), so an Ivy is likely to have similar or worse teaching than a community college. 2) There is a
569.
by blagie 3y ago
Web search says that contrary to stereotypes, the problem was that they did NOT have contractor staffing agencies, and that as a result, they were not withholding taxes. This was resolved by bringing in a staffing agency. A second problem w
570.
by blagie 3y ago
It's not the same. The Microsoft problem was * independent* contractors. I.E. treating people as self-employed. Normal contractors are employees of a temp firm. None of these issues apply there. Footnote: I started my career as an IC,

More ›