I know, I am in the same range of points and still asking questions has always been a bit scary.
I remember spending 2h writing a question for what I thought was a complex c++/compiler issue. 10s of thousands of lines proprietary codebase, so I couldn’t include everything obviously, but also couldn’t create a “minimal working example” to reproduce the issue. So I included as many things is I could to try to get pointer on how to track that behavior I was seeing. Of course the second I post it I got a -1 plus “can’t reproduce”/“please add minimal example”.
An other time, I had a question that was very similar to an existing one, but different setup and the answer did not solve my problem at all. Mentioned all that, linked the other question and specifically wrote that it was NOT addressing my problem. Posted it, soon after tagged as duplicate with that one answer that did NOT solve the problem.
After that I rarely asked questions again.
Also the points system made it frustrating as a new user: someone 2 years ago asks a basic language question “+50 upvotes”. You asked a similar question, asking extra clarification on an aspect “-2, already answered, read the doc” and so on. And with such a big deal made about reputation it felt like just being born early and being able to be an early adopter meant you got east points. For new users, though luck.
Skill issue to the second paragraph. You can always create a minimal example. I've done it many times. Some of them are quite long.
Start by copying your whole file with the problem into a new file. Delete each part. See if the problem occurs. If not, Ctrl-Z and try deleting the next part. Be extremely aggressive in deleting. Also try refactoring things to be simpler if deleting doesn't work.
In C++, inline #includes not from the standard library by copy-pasting the file contents. At each step, make sure the problem still occurs. Repeat the iterative deletion process.
You should now have a fairly short source file which exhibits the problem. Rename all proprietary identifiers to foo, bar, baz, frob. Submit question to SO.
The one advantage of actually trying to use SO was that the fear of asking a question usually made me do so much research that I'd solve my issue in the process of fully describing it.
That did also make the community lose out on the answer though.
Nuclear-powered ion thrusters could solve this issue. They provide low acceleration for a long time consuming very little consumable. This would allow the telescope to stay at the right position for observation.
Please oh please try to make Kärcher adopt your stuff. Even their latest premium cleaning bots are hopeless when they don't know where they are, even when I tell them where they are.
I have no hope for this working. I recently bought a dumb meatbag operated vacuum after tiring of the robot's issues. The constant getting stuck in places drove me crazy. After getting a new vac, I used it in a place where the robot had just "cleaned" that morning. According to the new collector's contents, the robot sucked at its job of sucking.
Mine worked pretty well as well and slightly improved after an update, but has gotten dumber over the years. Not only does it get lost, but it physically gets stuck under things that it did not used to get stuck under.
Can you please stop posting flamebait, unsubstantive comments, name-calling, shallow dismissals and the like? You've been doing these things repeatedly. It's not what this site is for, and destroys what it is for.
So the vacuum cleaner manufacturer would just require customers to install mistral on their local LLM hostbox? Or maybe they'd sell you a host to go with it you couldn't get into and honest cross fingers we aren't peeking!
Once prices settle down it doesn’t seem unreasonable that everyone who wants modern conveniences plops down $3k for a home computer.
It’s still cheaper than during the “home computer revolution” 80s/90s when normies were buying these at Sears and Radio Shack.
It would be nice to believe that could be possible, but people mostly switched to cheap laptops as soon as they became viable. No harm in being hopeful though.
True. But it would probably take a couple years to get spin up a new hardware product line anyway; by then local inference will probably be cheaper, right?
So, it has to rely on exact situational step by step commands? I'm wondering how one could conceivably deploy this in a useful way. Usually you'd need to mark areas on the map and then the robot knows where to go, A* is trivial around obstacles once you have that and a lidar. And lidars are an order of magnitude cheaper than something that can run an 8B VLA.
One could maybe autogenerate these text planning commands, but it would require a map and the robot's current location, so it doesn't really solve that, unless it can find a specific thing completely on its own. How much of a planning horizon does it have?
You probably don't need a geometric map. Just have someone wander around with a mobile app and feed the video into a more powerful model once, asking it to produce descriptions of the different areas of the office or building and how they connect. Now you have a "text adventure game" map you can use it with a small LLM to produce instructions for the robot to follow, assuming it knows where it currently is.
The advantage over traditional approaches is presumably flexibility. LIDAR isn't going to solve an instruction like "find the man with the pink shirt".
Yeah the benefits of a model that can understand text well enough is certainly that one could in theory just describe the layout. Just let it know what connects to what, what's on each floor, etc. Might work if it can figure out where it is based on contextual info, but probably not super well yet I imagine. VLAs are known to have very little text context, like sentences worth at most, maybe this one's different.
Do you think a similar model architecture could include some sort of memory, so a follow-up prompt of "Go back to the water foutain" would somehow work?
A in-model memory approach is probably still deep research but maybe a Rag-like pipeline could work in some instances
Could you please open source this or a 4B version? I’ve been messing around w hooking up vllms to cheap robots and skipping the whole ROS stack and this would be an absolute delight to play with
Cloud models can use batch processing which is significantly more efficient. A local model has basically a batch of one which takes as much time to process as a batch of 100 because the gpu is memory bound and spend most of its time loading the model from vram to the gpu cache while the gpu cores are idle. With a batch of 100 the model loading time and compute time are roughly similar. So local
Models have a first 100x lower efficiency. Secondly, local models are idle most of the time waiting for the user to write a prompt, so the efficiency gap is probably more around 1000x.
It's an interesting point but local gpu efficiency is not something I think about when I'm being rate limited or when my subscription costs keep rising.
I think folks in this thread are underestimating how expensive it is to serve a SoTA model at 100 tokens a second. In addition to the $500k in capital costs, you also have significant electricity costs.
This stuff is expensive because supply is much lower than demand. If everyone was to run their own hardware with a batch size of
1, we'd have 100x more demand for inference hardware and electricity than we do now, and people would be even more frustrated. Efficiency is everything, and we need all the economies of scale we can get to meet demand.
But that's why you shouldn't expect local models to provide quick real-time answers, at least not with the same smarts as SOTA models running in the cloud. Slow batched inference (if possible - RAM capacity can obviously be a challenge with typical models and end-user hardware) can be a lot more effective.
My point is that it is WAY more efficient if we put the world's DRAM supply into a shared inference pool instead of stranding it in local machines where it won't have as high of batch size or utilization.
The cost of not being efficient is even higher DRAM costs than we have now, given supply and demand.
Much of the world's DRAM stock is sitting idle in consumers' local machines and on-prem servers. If that DRAM gets some use, even "inefficiently", that's a meaningful decrease in demand.
That DRAM would get even more use if it was removed from these machines and placed into a shared pool :)
I joke, but thanks to the brutal DRAM market there has been some movement in this direction lately...
I think the question of who controls the model is far more pressing than the question of who owns the DRAM.
It's easy to rattle off a half-dozen different vectors of likely enshittification over the next few years -- ranging from increasing censorship, to lower rate limits, to removal of existing features and forced addition of unwelcome new ones, to extortionate price increases, to unexplained and irreversible account bans. The only way to avoid them all is by running weights you own on hardware you control.
How smart and how fast is your local model? Those are certainly important questions, but "Does it exist at all?" is more important.
There isn't enough hardware in the world for everyone to run their own SoTA model. The only hope we have is if we work together to host these on shared infrastructure, benefiting from >50x economies scale due to batching, etc. That infrastructure doesn't have to be owned by greedy corporations.
Yes but most designs still require metal bits. Typically the frame will be 3d printed and then you rely on machined components for things like slides and rails.
I have done that at meta/FAIR and it is published in the Llama 3 paper.
You usually start from a seed. It can be a randomly picked piece of website/code/image/table of contents/user generated data, and you prompt the model to generate data related to that seed.
After, you also need to pass the generated data through a series of verifiers to ensure quality.
For the knowledge preservation, I guess that a copy of deepseek has most of the required information. But, it would be hard to run it in a primitive world.
Thanks. I think the knowledge should contain at least text, image and tutorial videos. It would be nice if it contains some VR training programs too.
The only problem is, we need to find secure areas and long term power supply for these stuffs. The "good" thing about the virus is that it's not going to kill the infrastructure, but it kills the people who maintain them.
I'm thinking, maybe private citizens should prepare such a small library as well, just for survival. There must be some VR games about camping, fishing, tool making, etc.
But we have landfills which are full of great raw materials. I would argue that it is easier to collect steel from a landfill than from a mine during the industrial revolution.
Collect, yes. But those materials, at the end of the consumer use cycle, are definitionally about as far as possible from raw. Having collected this scrap steel, how do you propose to smelt it efficiently for reuse with little or nothing that burns hotter than the local hardwoods? That's the question actually being asked.
I believe you can work some grades of steel tolerably with manual, bellows-fired forge processes, but not all of it will still be the same kind of steel when you finish (decarburization, etc), some you won't be able to meaningfully hot work at all, and you likely won't be reliably able to produce pieces much beyond the quality you could get in bloomery days - forgings and hot working would benefit from well-chosen scrap of compatible metallurgy, but everything would tend over time to rehomogenization into something between wrought iron, and what we would now call low- to medium-carbon mild steel.
That's far from nothing, you can do at least as much with it as our ancestors did, but it also isn't close to anything we'd call "modern." Between the relatively
enormous energy inputs required to do any meaningful hot working and the relative scarcity of materials no longer being manufactured, tools and objects made of iron would probably come to be family heirlooms again for more than sentimental reasons: replacing your hipster thrice-great-grandfather's cast iron might indebt you the equivalent of half a year of your struggling truck farm's proceeds.
(It is still called a 'truck farm.' No one knows why. The old missus in town who reads says it has to do with some of the old machines, but even she doesn't try to pretend she ever saw one of them move, so no one thinks much of that. But all the village, not only the half or so she's midwifed, is happy to grant her her modest notions.)
"Not raw" is actually generally an advantage, as it's possible to sort through landfills to find directly-usable materials, or those which can be fed to an electric arc furnace (presuming that level of sophistication), or a charcoal-fed blast furnace.
This isn't great, mind you, but it's a good start.
Steel has a relatively high melting point. Other metals, notably aluminium and copper can be worked at far lower, far more attainable temperatures.
Keep in mind that modern steelmaking really doesn't begin until the Bessemer process (1860s), and that was far more predicated on high-volume, high-quality fuels (anthracite coal) than it was in the input iron ore grade. Knowledge of and access to liquid oxygen, far better process (and temperature) controls, and improved metallurgy, through the 19th, 20th, and 21st centuries have advanced smelting and fabrication even further.
NB: "Truck" means to barter or trade. A "truck farm" is one on which cash crops (rather than those for local consumption) are grown, usually vegetables rather than staple grains (wheat, maize, rice).
I asked no question for you to answer, save the one you actively and at not obviously needful length failed to treat at all. How do you propose to power an electric arc furnace on peat, coal, or hardwood? How do you propose to produce liquid oxygen with like premodern fuels, which the problem statement declares as a constraint? Save the trivial point that less useful metals with lower melting points are easier to work, what does any of what you've said have to do with anything I and my prior interlocutor actually were conversing about?
That last question, though at least actually asked, also requires no response. Its answer is obvious, also trivial, and unrelievedly in the negative.
I suspect you and I share similar pessimistic views of what a post-industrial, post-carbon, post-collapse scenario might be capable of. I was amending and supplying technical corrections to your points, though I agree generally that landfills won't come close to replacing historical raw material sources.
Addressing the question of metalworking and presuming a rubbish tip source, it's useful to note:
- Previously refined metals won't need refining or smelting but rather recycling. This generally simply requires heat. Aluminium does not, for example, require (electrically-driven) reduction, iron and steel do not require (coke, oxygen, or hydrogen based) carburation. The heat requirements remain challenging, but it is still a lesser challenge than production from ore.
- There are conceivable thermal processes which might assist, and for which technological knowledge even in the absence of prodigious energy resources might suffice. Solar thermal energy (requiring polished mirrors, but these being reasonably attainable) can achieve temperatures of 3,500 °C (6,330 °F). Steelmaking "only" requires temperatures of ~1,700 °C. Total capacity of a solar furnace would be well below that of a fuel- or electrically-powered blast or arc furnace, but useful quantities of metals and glass could likely be produced without extirpation of forests for charcoal.
- Siting near geothermal or hydroelectric resources, and presuming electrical generation, could enable electric-arc furnaces. Even today much aluminium production is opportunistically sited near such cheap power sources.
Several of these options might not be immediately available following a widespread collapse, but could be bootstrapped within reason over time, though most likely at far lesser scales than at present.
My suspicion is that a post-collapse society, and/or a future technological society operating with a rewewable energy basis (biomass, hydroelectric, geothermal, solar, wind) would probably have very different material bases (far more stone, brick, and ceramic, some plant-based materials whether structural timber or plant-fibre-reinforced ceramics), a vastly different land-use pattern (concentrated rather than sprawled settlements) and transportation (water-borne, canals, heavy reliance on pedestrian travel, possibly electric-powered transit and freight, minimal air travel). Overall energy-intensity comparable to the late 19th / early 20th century in the US and Europe may be reasonably attainable with smaller populations, and on balance that wasn't particularly burdensome. Food production is probably the major hurdle without Haber-Bosch ammonia production. Sufficient farm mechanisation given some available fuel and/or power distribution (alcohol, biodiesel, possibly synthetic hydrocarbon production, elecricity) would have a huge societal benefit even at small fractions of present total and per capita energy usage, not just for agriculture but other high-benefit uses such as marine propulsion and remote heat and power.
This is a discussion site, and the discussions occur between more than just two people (e.g., comment and response). I was responding with additional context for any reader interested, though of course you can be presumed to have interest. Discussions are not however proprietary, and shouldn't be treated as such.
