this post was submitted on 01 Oct 2023
1140 points (97.5% liked)
Technology
59703 readers
5399 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
Ok, hear me out: crypto, based on "proof of training an AI"
If it takes so much power, it must be secure, and this way it wouldn't be "totally wasted"...
I'm not sure if you're serious, but just in case: that wouldn't work, mining is really just verifying transactions. So if you're not doing that, you may earn crypto by "mining", but you can't spend it because no-one is verifying your transactions.
Not correct.
Mining is a "proof of work", in the case of Bitcoin it's competing to be the first to find a hash that meets certain parameters (difficulty), for a block referencing the previous top one. Whether the new block has transactions in it or not, you get the same reward for being the first one to find and broadcast it.
Verifying is done by every node in the P2P network, both when deciding whether to relay candidate transactions, and when checking whether a new block's hash meets the mining requirements.
The Bitcoin blockchain has plenty of valid blocks with no transactions in them (part of a speculative mining strategy used by some to get the block reward faster than others).
The whole scheme works the same with any other kind of "proof of work", as long as the nodes relaying the new block can check whether the work happened or not (there are many ways in which that could be accomplished for AI training, the easiest of them by publishing the new model and having nodes check whether it meets some quality parameters).
I mean, yeah. I knew most of that, but I just wanted to keep it short and simple.
I don't really understand how it would work with AI training. If your computers are working on training AI instead of finding blocks, I don't see how you can support transactions. Just sounds like distributed computing with rewards to me, where you might be able to cash out at some central portal or smth, but you can't send other people that money directly (at least not over a blockchain, but would be possible vis that portal maybe, although, again, that wouldn't be a blockchain).
Proof of work need not be useless. E.g. https://primecoin.io/
The tricky bit is finding a problem that is hard to solve but easy to verify. I'm not sure AI tasks fall into that category.
The transaction verification is separate to the work.
They actually do. Training an AI involves changing some values in the model in an attempt for it to better fit an optimization function. It takes many tries to find a set of values that perform better, but a single try to confirm it does.
Both sides require much more computing power than for a single hash, but the difficulty imbalance is still there, and verifiers could change "how much better fit" the next model needs to be, just like they do by changing difficulty requirements right now.
True. The next iteration doesn't need to be optimal, just an improvement in the loss function.
Not sure how they would decide when to stop.
Huh, neat. Thanks for the info!
There is no "finding blocks" in Bitcoin, it isn't Minecraft. Miners work on "finding a better hash", for whatever block they want to propose. The two actions, creating a block, and working on finding a hash, are separate.
In a "proof of training an AI" blockchain, there would still be a hash linking one block to the previous one, just the proof for accepting a new block would no longer be looking for another (useless) hash.