It's wild to me that PoW systems are how we sort the bot problem... the bummer is that all this "work" is just wasted cycles, at least in crypto there is a token you can sell.
Scraping would be costly in this world: scrapers would have to spend tokens in order to get the webpage.
But, in this world, the website owner would receive tokens that they can then use to do whatever they want, including paying for servers and bandwidth. This is the sense in which the cycles aren’t wasted: the website owner now has cash to spend.
Effectively, both scrapers and ordinary users would be paying for the privilege of getting website bytes.
This also solves the problem of having to wait for your phone to solve the challenge while you’re browsing: you can buy or mine some tokens ahead of time and pay them as soon as challenged. So can the scrapers, but because they’re accessing enormous numbers of pages it’s hopefully prohibitively expensive for them.
Yup was thinking the same: make honest people pay $0.00001 when they visit the site (in electricity/compute), have the challenge made so that only the website wins a tiny something. Bleed the bots dry.
> This makes Anubis challenges use a memory-hard proof of work function (argon2id) instead of just a CPU hard one. It also means that the "hey Claude vibeslop me a CUDA Anubis solver" route is on its way to being fundamentally dead.
I want to do this eventually, but it's hard to split things into the micro-tasks that would be required to make this work on Anubis. One of the ideas I'm throwing around is a world where Anubis helps fuzz old games to find timesaves in tool-assisted speedruns. It's harder than you think.
The tools that exist for fuzzing speedruns (TASers call it "botting") are fairly primitive, excluding a few game-specific ones. Because of that, there really aren't that many TASes where a distributed randomised search could make improvements which don't get immediately overshadowed within a day of human attention. Improving botting tools and increasing their adoption would be more effective IMO. (Incidentally, TASVideos.org recently banned one of the people pioneering bot development.)
Meanwhile BOINC is well-established as the platform for distributed computation. If you can figure out how to squeeze its work units into Wasm challenges, I'm sure a lot of researchers would thank you.
BOINC isn't really built for "small tasks", afaik. Once you get small enough, the work the server is doing to manage the tasks gets to be around the same cost as the server just doing the tasks itself.
Yeah, I'm using PoW-based "DDOS defense mechanism" for a current project and had this thought too. I briefly looked into what it would take to exploit PoW to do something economically valuable like folding@home, and it looked a bit tricky. At first glance, these sort of projects seem a bit unsuitable for real-time PoW because they were designed to be run on desktops and so tend to do things in large batches. There probably are other use-cases that could be more suitable though.
Might be worth chucking that thought into Astra, especially if someone springs the $$$ money for it?
Transparency logs. It's mandatory for a cert to be in CT logs for browsers to trust it. Those are public, if this was happening, someone would have noticed already.
I don’t think many realize how hard it is to raise for a no revenue company that isn’t buying out large GPU contracts and racking machines.
They have a winning product for sure… the worlds models have chosen a hugging face workflow, but honestly where would they be able to find the revenue required to justify another couple rounds at ever higher valuations.
To me, I think these folks see that the music is getting harder to hear and might stop entirely, might as well cash out. I do hope their employees were well taken care of, and 13b of NVDA is likely to be worth a lot more in the coming years.
Agreed. Take a ~$13bn buyout im sure cash and stock is the smartest thing to do for a company their size and age. Take the profits now and keep building securely. I imagine their employees very much gotten taken care of. Jensen mentioned an additional $1bn in cash compensation etc to keep employees at HF.
> To me, I think these folks see that the music is getting harder to hear and might stop entirely
At the frontier companies, but not for hugging face or nvidia. Their business are going to be around for a while considering how local models are performing now.
I don't think they have much data, they host the (open) models but I don't think they're getting any logs from users or anything like that. I don't believe they monetize data either. I also suspect Nvidia has a much larger audience than Hugging Face.
Revenue is the small part of it though. Every training and inference script in the ecosystem imports their client library and pulls from their hub by default, so they sit on the default path between a model and the machine that runs it. For company that sells the machines, owning that default is worth more than any subscription line.
I don't think local models would require hardware at nearly the scale that the datacenter buildout for OpenAI/Anthropic/etc. required. Nvidia got as high as the Icarus, it will be hard not to melt, despite they being a real company with real product that is so in demand that you can't buy their hardware even when you're willing to pay above the recommended price. They are addicted to billions, and the cold turkey clean-up will be painful as hell.
ColdFusion is the COBOL of web programming… there is some human at an insurance/mortgage servicing company thats not had to learn a new tag in 5 years pulling in 200k.
Its really awesome to see large organizations like Alphabet/Google doing good work for the general betterment of cryptography. Tools like these are unique to our time, provacy concerns have never had such well capitalized adversaries.
reply