I can confirm through my family’s history that coming of age in that era was also life on easy mode for very-poor white men who moved to the city early and were willing to able to show up to work consistently. No degree needed. Everyone who fit that description did really damn well, often despite multiple mistakes the presence of which in a bio would mean fucked finances for life for someone today.
Pensions were magical and housing was cheap. Powerful combo.
Unemployment insurance is a form of insurance in US that pays you if you lose your job due to layoffs and other conditions. It is paid by you so it is not “welfare.” It is managed by each state independently and has some rules. For example, if you lose you job today and “don’t claim” unemployment insurance for two weeks, those two weeks are not paid out to you (this is in Texas). Every week, you must show that you are “actively” looking for a job. You may lose the benefit if you are not, or if a job is available but you decide not to take it. You must report on your efforts to find a job (interviews etc), and must attend some state class about looking for work.
Most states’ unemployment insurance covers up to a maximum benefit per week, and a maximum number of weeks. When I lost my job once, I was paid $325 per week.
At least in my state, the employee does not pay unemployment insurance tax, but the amount the employer pays is usually reported on the paystub. That is, if you are paid $85,000 salary it won't be reduced by the unemployment tax. However, some economists treat all payroll taxes as effectively paid by the employee out of "total compensation".
Given how much you liked hanging with ants, I recommend you try spiders in a garden. Look for a spider web where one is hanging, toss a small piece of debris that breaks a little of the web, and watch in awe how they repair it.
Haha. Not a huge fan of spiders, if I see one in my home I’ll catch him and put him outside.
Idk why but I feel really bad for critters. I’m in SF next to a park.
Every morning on my walk if I see a pill bug, a worm or caterpillar, or some other critter on the side walk I’ll grab a leaf and put them back in the park. Don’t want them to get squished :(
Back to spiders tho. I had jumping spiders in my last apartment. Had a really pretty patterned one in my office that would turn and face me whenever I moved. It lived in my blinds for a few weeks and then I never saw it again :(
Had some pet silverfish too. I have a TON of moleskines and I’d feed them old pages.
Sorry I’m probably weird for loving all these little “pests” but I find them really wonderful.
I rarely go through this part of the basement, I spilled about a table spoon of water on the floor and I come back 10 minutes later and a giant 7cm leg span Wolf spider was drinking from the newly formed puddle. Now I make sure there is plenty of water available in the basement.
I would hunt for my friend spider. Find an ant, throw it into the net, watch how quickly spider friend descends right to the spot and spins it into a take-out container.
Really only works with live victims. If they died before I threw them on the web, the spiders usually wouldn’t recognize from the vibrations that there was food at that grid location. I did trick a spider once with a pine needle, tapping/shaking the body of the dead prey until I got lucky and the spider came. Usually this didn’t work though, tripping whatever heuristic the spiders used was tricky.
As an ignorant person but one who has been trying to figure out if KGs are superseded by LLMs, is there a source you can think of to figure out if both have a place in an architecture, or if LLMs are sufficient? What sort of use cases require both?
Current performant (high accuracy) LLMs have a quadratic cost (space and time) in sequence length, they have a finite size typically much less than any KG of note, their connections are all fuzzy, they aren't especially amenable to small updates, training and fine-tuning don't work well with high-entropy data, and most computations physically cannot be done via any single pass through an LLM regardless of how it was trained.
Those constraints together create a landscape where if you have a big knowledge graph there will invariably be important questions the LLM cannot appropriately answer about it, no matter which strategy you use to try to ramrod the KG into an LLM architecture. If you don't train/fine-tune the LLM on the KG, it doesn't have enough context to answer your questions. If you do, your KG doesn't have enough data duplication to allow training to work well. If you manage to train it anyway, you can't ask compounded questions because of the max LLM circuit depth. If you try anyway and just run the results back into the LLM as input you have a compounding error effect because the whole thing is fuzzy. If you try to circumvent that with error-reduction techniques you tend to blow through the current context windows (quadratic costs) and still have unreliable results. And so on.
None of that is necessarily true forever, but suppose you have a problem where a KG is a natural fit but some of the data is a little fuzzy (you have pretty good graphs of how cities and roads and individuals and companies and whatnot are related, but it's not perfect, and some of it is textual or not otherwise appropriately structured). The KG can answer a number of queries very well, limited by the lack of structure in the node/edge representations. An LLM can't do much because it can't compress all those possible edges into its weights, because it can't fit the whole KG in a context window, it can't be appropriately fine-tuned to the data, and even if it could it couldn't recurse well without compounding error. If you instead use the LLM as a pre-processing step on the nodes or as a fuzzy neighbor search (restricted by the KG) or in some other way, you get a data structure that looks a lot more like a better prepared clean KG and can run traditional KG algorithms to ask questions like who might need your tax prep services or whatever. Getting an LLM to do that for the same cost will take a _ton_ of engineering beyond what I've seen poured into the space.
Materializing the attention matrix has you slower than FlashAttention much of the time, just because you get severely memory bandwidth bound and are worse off than mild extra computations resulting from streaming it (and especially re-computing the attention for the backwards pass, though that won't need extra memory bandwidth, just minor compute.).
I am not talking of managed services, but infrastructure services.
This feels like a Japanese soldier in the jungle still shooting left over ammo after the war has been over for more than 10 years.