pbmonster
5d ago
It can go a long way by basically feeding you an existing project and not mentioning where it got it. I don’t mean to oversimplify and say that this is all models do and that there’s no reasoning, but I think people underestimate just how much is memorised
I have the same experience. It's just so glaringly obvious when it has to code something that is rare in its training data.
Load this folder of data (convoluted structure, some data is binary): perfect oneshot
Build a database from it: perfect oneshot
Do some signal processing on the data for cleanup: good ideas, uses the correct libraries and algorithms unprompted, but not immediately usable
Make plots for me to debug the signal processing: perfect oneshot, signal processing is now usable with minimal handholding
Data analysis, involving multiple integrals over the data along different axis: completely wrong, not even close. A first year grad student could solve this in a few hours, this is in at least a dozen of standard sub-field specific text books and has a very specific name - but it's probably not on github.
Discussion about a factor of 2 included in some text books, but not in others: completely wrong. having this discussion with an actual parrot would be equally as productive as the stochastic parrot.
pbmonster
7d ago
Here’s why it’s a problem. (Some basic science)
Slopslopslop.
And not even good slop, no, non-thinking low parameter slop.
pbmonster
12d ago
That's insane, is this online shopping from unknown brands or emergency shopping right before an event?
My hit rate is 100%, because I'm picky and I only shop in person and I try everything on. I often leave without buying anything, because I know what fits and what looks work for my body type and aesthetic.
Either you need to simply walk away more often, or this is a skill issue. The latter can be condensed into a hand full of straight forward rules. "The cuff needs to be this long", "the shoulder seam needs to sit right on this bone", "lifting both arms must look just so", "squatting in new jeans must feel like that and show this much ankle". The exact values for all that can be gotten from YouTube or guides on manosphere blogs and /r/malefashionadvice. All 3 are dangerous, because half those people dress like retards, so look at the pictures and decide if you want to look like that.
I've added a couple of additional rules that save me lots of time by not even trying things on ("absolutely no synthetics, 1% is unacceptable", "the maximum size of a logo is 1/2", "no baggy/skinny cuts", "only plain, neutral colors"), but that's basically it.
pbmonster
12d ago
I mean sure, there's tons of jobs in robotics like that, but those robotics competitions were designed (at least originally) to funnel young nerds torwarts going to college for electrical engineering. Because the field needs large amounts of those, too, and for the last 20 years you could more easily make more money (unless you made it into Boston dynamics et. al) if you went into pure software instead. So a little PR for a career close to the hardware was a good idea in principle.
If you'd want more technicians straight out of school, you'd do something like an RC racing league with a build phase for the car/drone.
given that the birth rate decline is being caused by less coupling, and not by mothers having fewer children.
It's all of those, right? There's women who want children but lack partners; but there's also women who have stable partners and never end up having children anyway; and women who have children, but stop after 1/2 - either because they started to late, or because of lifestyle choices (housing/career/hobbies being in conflict with having a larger family). I suspect each of those groups to be significant. "Conscription" could help with the latter two.
Of course, conscripting women for motherhood is gonna be tough
I think there's pragmatic ways. Allow women to delay their non-military national service until they're 25, and wave it if they have children by then. Hell, do the same for men. Allow the women to drop out immediately upon reaching the second trimester while still maintaining their pay. Have the non-military national service use mandatory barracks for unmarried personell, and co-locate those with the military bases that do basic training for the conscripts. Don't actively hinder the inevitable parties to much.
Still, no way this gets you TFR > 2. That requires a pretty radical social change. I suspect making parenthood high status might be the only way.
pbmonster
20d ago
I don't think I have ever, in my entire life, heard anyone discuss fertility rates "in real life", outside the Internet.
Interesting, is this in the US? In Europe in general, but especially in France and Germany, the "demographic transition" has been a permanent fixture in the media and in public discourse for at least 30 years. I have childhood memories of seeing inverted age pyramids in newspapers.
Understandably, since the fertility rate (or "children per family", the more palatable euphemism) has cratered earlier here, and the financial scheme behind their social systems is reliant on young workers. Which means they are even more fucked than all the other countries with low fertility, so at least they are aware of the problem.
pbmonster
21d ago
And if they don't die, they need to change. The doctors know the stats, at the very least they could/should be frank with their patients.
A friend of mine was told during her first consult with the IVF clinic that a chance of success at "high confidence" would require a number of eggs equal to her age at implantation - so to prepare for 3 rounds of egg retrievals at the bare minimum, and as soon as possible. She got unlucky, and the first round only retrieved about 4, so the number of cycles was immediately upped. When she asked if they couldn't try those 4 first before cycling again, she was advised to not waste time and get the inventory as young/soon as possible, and to expect more setbacks.
Sounds like this lady did a single (more successful) retrieval cycle, and nobody showed her the math.
pbmonster
26d ago
OK, all good arguments. Maybe I'll have to take a deep dive on what's possible in open source land on that front (I refuse to touch Adobe/Abelton et al.). Do you use commercial models/implementations or do you have a recommendation where to start if I want to set it up myself?
Not to mention they tend to have ridiculous censorship
That's the nice thing about just renting a VM on expensive hardware from runpod. They don't care at all what you do, because you provide all the software yourself.
pbmonster
26d ago
Interesting, does the software integrating those non-LLM AI functionalities offer to run this in the cloud for you? My experience with both image generation/manipulation and local LLMs has been that its almost always better to run those loads in the cloud - either directly from the big AI labs, through a vendor like openrouter.ai or on a rented GPU like runpod.io.
You can very well do it all locally, but it's a pointless toy with 8GB of VRAM, semi-interesting with 16GB and you're finally cloud-independent with 24 GB of VRAM. And you can get many hours of GPU time on runpod.io for the price you'd have to pay for this much VRAM.
But yeah, I haven't worked with audio models and the things Photoshop can throw at a GPU.
pbmonster
26d ago
what stupid and crazy people are like. They always existed, and always had these ideas. The only thing that has changed is that Duggan can now see them. Nothing got worse - if anything from my impression, the internet makes correct information easier to get than ever
I think unfortunately, this also goes the other way. Misinformation is also much easier to get, giving us the toaster fucker problem:
"Man wakes up in 1980, tells his friends "I want to fuck a toaster" Friends quite rightly berate and laugh at him, guy deals with it, maybe gets some therapy and goes on a bit better adjusted.
Guy in 2021 tells his friends that he wants to fuck a toaster, gets laughed at, immediately jumps on facebook and finds "Toaster Fucker Support group" where he reads that he's actually oppressed and he needs to cut out everyone around him and should only listen to his fellow toaster fuckers."
I have previously been using an old non-subscription version of AutoCAD, but I’m not sure the budget will support paying for a subscription moving forward, so I’ll need to switch to something free/cheaper. I was thinking most likely FreeCAD.
This sounds like light hobbyist use, like for some 3D printing? Any laptop from the last 5-7 years will have no trouble running that. Unless you're designing large assemblies or planning on doing simulation work on your designs, I'd just get >16GB of RAM and call it good.
I'm a firm believer that nothing important has happened in the PC space for the last 15 years (Apple Silicon is another matter, but it didn't sound like you're interested in that). We did everything you want to do 15 years ago on Windows 7, and it was fine. Because we suck at software design, doing the same thing now needs more RAM, so we gotta buy that. Done.
If my current PC dies, I'm getting another Thinkpad X1 Carbon, probably 8th Gen (those are from 2020 and go for around $500 when getting a good condition business lease return, and sold for >$2000 6 years ago). Matte full HD screen, good keyboard, solid build. X series Thinkpads have lasted more than 10 years of hard use under my care historically, and I would expect this one to continue the legacy.
But really, any PC from that era will do fine at your workload. I'd choose on price or other secondary features. If you get a desktop tower, spend your money on a nice display/mouse/mechanical keyboard and maybe passive cooling instead of compute.
pbmonster
26d ago
The writing smacks of the fedora, if you catch my drift
Reddit certainly loves it, yes. Read it on recommendation from one of the SF subs, and hated it. Its main weakness is that Locke is such a Mary Sue. I find characters that simply excel at everything annoying. Other than that, it's decently written Fantasy James Bond with forgettable world building.
Would not recommend.
do slave concubines even count against the 4 wife limit?
Of course not, what are you talking about!?
In that sense, the predominant genetics of the area got polluted.
There's no genetic evidence of this happening. Northern and Southern Italians differ genetically quite significantly, but that difference goes back much longer - bronze age and copper age. Southerners are most similar to Greeks, while northerners are most similar to French.
Also, that span of genetic difference isn't really that uncommon. North and South Germans have similar genetic differences, for example.
Maybe it's a regional thing?
At least internationally, for sure. Ironically, women in France almost never wear "French manicure", but anecdotally I'd say wearing nail polish in a muted color on short nails is more common than in the anglosphere. Practically all French (or western European women in general) I can think of at least sometimes paint their nails at home, and wouldn't even get acrylics for their wedding.
pbmonster
1mo ago
You go on to propose that black men should be encouraged to be less masculine, basically. In other words you are trying to bring him down to a level where you can compete with him as you are a non-black man.
submit to the stark reality that women choose mates based on primal instinct of masculinity and dominance and physical power which you or the less dominant man knows he can't compete at and win.
Come on, this is oversimplified to a degree it's wrong. "Masculinity and dominance and physical power" is only one dimension of what makes an attractive man. There's also charisma, verbal/emotional intelligence, appearance/style, competence, and (of course) status.
If you send the most masculine gangbanger to a party also attended by a sunny boy pro surfer, a lead singer/guitarist of a half-decent band, a frat bro lacrosse player, a male ballet dancer and a navy fighter pilot - and they all agree that gangbanging is low status - the gangbanger will not pull, not even secretly.
pbmonster
1mo ago
I don't just want an open source model roleplaying as Homer Simpson (or whatever), I want it thinking nothing but Homer thoughts from the ground up.
The main problem for something like that will training data. You say "at only minor expense", which means you're considering renting GPUs (as opposed to just training on a gaming PC)? The most common tutorials on that scale use many hundreds of millions of tokens of training data (training on TinyStories (500M tokens) and Wikipedia English (4B tokens) is common, books3 (100B tokens) is sometimes used for more ambitious and capable toy models).
And even if you just take a small open source model (or your own toy model trained on TinyStories) and do some post-training fine tuning, you'll still need millions of tokens of training data. I'm not sure Homer had enough lines across the almost 40 seasons to even get this far.
And all but one of male friends have paired up with women who are also not like this, even if some may be a little woke, and only one or two ever spent significant time on the apps the rest just met at work or school like the old days.
Exactly, and all my female friends and family members also all have paired up with entirely normal men (desperately holding out for Chad to finally commit never seems to have been an important part of their lives). The vast majority also still get married.
With regards to the fertility crisis, the real problem seems to be that all those "normal, old school" couples seem to consciously choose to have exactly zero or one child. Even two children is rare in my bubble, three or more totally unheard of. Which, of course, results in exactly the TFR we're seeing.
I don't really understand why. Especially the in-official one-child-policy baffles me. Maybe having one child already scratches the itch of parenthood enough? Maybe the tolerance for discomfort has gotten lower, and most young parents end up being surprised how difficult (especially without significant support from family, who universally live far away now) the first few years are - and choose never to do that again? Maybe the "universal theory of housing" is correct: most of us live in classic 3 Bedroom apartments, condos or town houses, which works out exactly to parent's bedroom + child's bedroom + study/guestroom?
In the era of ChatGPT, Google, and Wikipedia, ignorance is inherently willful.
Don't overestimate the median poster on social media, especially on the images-and-video side of the internet. Even GPT is TL;DR. Wikipedia isn't only much to wordy, it's straight incomprehensible. Google just wants to sell them something, or get them to watch a video made by somebody just as well informed. I'm not saying it doesn't often start with a bad actor working on an agenda. But the spread (and memetic drift) is entirely fueled by very incomplete information supporting a preexisting ideology.
I've met well-educated and otherwise capable people who support "de-growth" and "solar punk" aesthetics. Their eyes glaze over when they hear the first "calories per acre" or "supply chain of poly-silicon". They like the idea of gardening. And that's enough.
pbmonster
1mo ago
I don't know how common that's in the US, but doctors at hospitals are probably the career professionals that can most easily reduce their working hours without significantly impacting the quality of their work. Want to only work 2 days a week? No problem, just take fewer shifts, and the hospital hires more doctors of your specialty.
In Europe, this leads to many doctors reducing working days while their children are small.
the people who want intelligent reasoned discussion are the same people who come to the correct conclusions on most topics.
"Most" is doing a lot of work here. The audience of theMotte is only human, and as such perfectly capable of dealing with cognitive dissonance in unproductive manners. From the top of my head, I remember several discussions on "leftist" topics like vegetarianism, anti-car ideas, electric cars, solar power, urbanism, ect. that not only didn't result in "correct conclusions" (whatever that means in any context), discussion (and voting) wasn't very "intelligent" or "reasoned".
I think you could discuss those matters objectively and rationally. I think on a topic of... I don't know, "the culture of meat consumption in the west, and the necessity of factory farming to sustain it" you could come to a objectively and rationally "correct conclusion" after considering the ethics, economics, health impacts and negative externalities.
But I also think you couldn't do that here.
That's just classic "toxic masculinity", with "macho" and "chauvinist" being the closest linked slurs.
They don't land as hard, maybe because of the tacit acknowledgment that many women find some degree of macho/machismo attractive, and "chauvinist" is just to... intellectual, I guess.
pbmonster
1mo ago
Having the LLM do expensive things cheaply in e.g. python is doubly beneficial, since a lot of the tool calls (again: game states, chess engines, navigation software, ect.) are not only much cheaper (in FLOP/s), but also are often CPU loads instead of GPU loads. That frees up the GPUs for LLM inference of other customers, while a cheap CPU runs some python scripts on the side.
An again, the results will be much better. Using python for math will get correct results to math questions. Calling a chess engine means almost all humans will loose that game. Only calling a map app unlocks correct world model answers to navigation and spatial reasoning questions.
pbmonster
1mo ago
Alright, I stand corrected. Still, sounds like it was a lot less frequent for agricultural labor (which probably always was the vast majority of slaves) than in other systems of cattle slavery.
pbmonster
1mo ago
Yes, all of those. Instead of burning hundreds/thousands of tokens on game state/memory each turn, you can run a simple state machine for comparatively free compute. This will also be more deterministically correct (you could even do correctness proofs if you really want to, or just have the LLM write a bunch of tests), because if you keep game state as tokens, you currently have no guarantees/tests.
This is optimistic. Land borders are irrelevant, the Caspian is irrelevant: because we're talking multi million barrels per day, the only thing that matters is existing pipeline capacity, available short term. And there's really only three pipelines in the area that matter: the Saudi East-West pipeline to the Red Sea, the Abu Dhabi Crude Oil Pipeline to Oman, and the Iraq-Turkey pipeline to the Med. Their max capacities are around 7 million barrels per day, 1.5 mb/d and 1 mb/d, respectively. But they all already carried fuel before Hormuz was closed (albeit not at full capacity), so if you want these pipelines to substitute for crude carriers through the strait, you might get an additional 4 mb/d through them, 7 mb/d if you're lucky (that requires the Saudi to do some real magic to their pumping stations, and you need to unfuck the political trouble between Iraq and Turkey that prevents them from using that pipeline - and even then, it doesn't even connect to Iraq's major oil fields in the South).
That leaves the global market short 13 mb/d, -13% in total. And that's a major oil crisis, almost double the shortfall of the Arab oil embargo of '73, which quadrupled oil prices. Once the strategic reserves run dry, the oil price is going to do violent things.
And then there's, of course, the matter that pipelines are extremely large, effectively impossible to protect and thus really easy to bomb with drones.
More options
Context Copy link