Basically a deer with a human face. Despite probably being some sort of magical nature spirit, his interests are primarily in technology and politics and science fiction.
Spent many years on Reddit before joining the Threadiverse as well.
You are speaking of “model collapse”, I take it? That doesn’t happen in the real world with properly generated and curated synthetic data. Model collapse has only been demonstrated in highly artificial circumstances where many generations of model were “bred” exclusively on the outputs of previous generations, without the sort of curation and blend of additional new data that real-world models are trained with.
There is no sign that we are at “the peak” of AI development yet.
Except it’s not denying service, so it’s just a D.
And yet the synthetic training data works, and models trained on it continue scoring higher on the benchmarks than ones trained on raw Internet data. Claim what you want about it, the results speak louder.
Yeah, AI has become good enough at this point that you can provide it with a large blob of context material - such as API documentation, source code, etc. - and then have it come up with its own questions and answers about it to create a corpus of “synthetic data” to train on. And you can fine-tune the synthetic data to fit the format and style that you want, such as telling it not to be snarky or passive-aggressive or whatever.
Stack Overflow has been toxic for a long time already. It’s one of the things that a lot of people seem pleased to see AI devour.
I have little to say about this because it’s basically nonsense.
The top 100 accounts already own 35% of the network and rising … that would give Lido and Coinbase a combined 51% of the stake
These two statements are completely incompatible with each other, for example. Also, the top 100 accounts only holding 35% of the network is remarkably good. And Lido is not controlled by a single individual or organization. And holding 51% of the stake means nothing on Ethereum, it works differently from Bitcoin. And the Gini coefficient is something that applies to national economies, not to blockchains. And I could go on, but this is just nonsense and you clearly have no idea what you’re talking about.
There are lots of uses that don’t depend on external data. The general “digital payments” use case, for example, just requires tokens to be transferred between two accounts on the blockchain itself. People then make use of that information externally however they like.
There are also ways to get external data onto the blockchain in a manner that is reliable enough for some particular purpose or another. A lot of game theory goes into that sort of application. Stabletokens have price oracles, there are prediction markets, “proof of personhood” protocols, etc.
The important things to consider are:
This applies to pretty much every field of technology.
Admit what? There are a huge range of possible designs that could be called “Proof of Stake”, and some of them could easily have that flaw. You can always design something poorly. Ethereum, the most widely used Proof of Stake token, doesn’t have that flaw in its design.
Modern blockchains like Ethereum don’t use ASICs and don’t consume any more energy than a normal database would. They use proof-of-stake rather than proof-of-work.
There are a type of tokens called stablecoins that are specifically designed to have a consistent price over time. Use those for your store checkouts.
There is one core purpose that blockchains solve that traditional databases don’t; decentralization.
If you’re running something that can work just fine with a centralized database, go ahead and do that. Use the right tool for the job.
Blockchains are for applications where you can’t depend on a “trusted authority” to support you. When you want to be sure you’re free of any outside interference in the things you’re doing, that no server can “go down” and take all of your data or operations with it.
An example of a non-financial-related activity I can offer is the Ethereum Name Service, ENS, which is a decentralized version of the DNS system. You can register a domain name with it and never have to worry about it being “seized” or going offline due to a networking failure.
You have to solve the sibyl problem, or else someone can just run 10,000 copies of the validator software on one computer,
That’s solved on Ethereum by requiring you to stake tokens that cost money. You would need an enormous amount of money to afford to spin up 10000 validators.
submit enough votes for a false record that it overwhelms any competing votes
I’m afraid you’re not very familiar with how Ethereum works. 10,000 validators isn’t anywhere near enough to disrupt the system, all you would do is burn your stake and lose all that money if you tried that.
Even if you acquired enough stake to prevent finality - 2/3 of the total stake would be required, costing tens of billions of dollars and taking years to work your way through the entry queue - all you’d do then is cause a huge annoyance to everyone on the system while your tens of billions rapidly burned down to below the threshold and finality resumed again. You wouldn’t be able to insert “fake” transactions.
People have been working on blockchain technology for a long time, these sorts of basic attacks have long ago been accounted for.
Which is completely useless if the data that goes into that blockchain isn’t guaranteed to be correct.
You’re missing the point of this. It just matters that it’s consistent. The transactions that are put into the blockchain just have to follow the rules of the blockchain and be the same for everyone who reads the blockchain. That’s all that “correctness” means as far as the blockchain is concerned.
And if there’s money in it you can bet your ass that someone will try to game the system.
Blockchains depend on everyone involved trying to game the system. They’re built using game theory to ensure that the most selfish actions for any particular actor are the best ones for the blockchain as a whole.
This isn’t anything like how validating in proof-of-stake works (at least not in Ethereum’s version, which is the one I’m familiar with). Validating is permissionless. Anyone can set up a validating account, there’s no authority that can reject you.
Who “authorizes” keys to determine if a particular solarpunk miner is sufficiently solarpunk?
No, that’s not how proof of stake works. Or it doesn’t have to, at any rate. Ethereum’s staking token is not a governance token, holding a lot of it doesn’t give you any “ownership” of the blockchain as a whole.
Quite the opposite in fact, if you’re staking millions of dollars worth of tokens then that means the blockchain has millions of dollars worth of your assets “held hostage” to ensure you follow the blockchain’s rules. If you don’t then the hostage gets slashed.
It’s not about the Fediverse’s incentives. It’d be the incentives of the bot-runners that makes them run bots. Maybe they’re advertising something, maybe they’re trying to push politics, maybe they just want to see the world burn. They will be able to run bots on the Fediverse to do that.
Yes, that’s me, and it’s not blaming. It’s describing the problem. I’m not saying whether these things should be law or regulation or whatever, I’m explaining why these things are being overturned in court.
I’m really baffled by what the point of this argument is. Are you trying to say that things are working as intended?
I’m not “blaming” anyone. I’m describing what’s happening. The reason why it’s happening is irrelevant.
That leads right back to the first thing I said in this comment thread:
And once the Fediverse is big enough to be relevant the bots will come here too.
It’s too small to bother with right now.
If we truly get to the scenario where it’s impossible to tell bots and humans apart, even while considering post history and everything…
We have the technology for that scenario right now, it just hasn’t been implemented on the Fediverse yet (as far as I’m aware).
I’ve done a lot of playing around with locally-run LLMs, and they’re quite good at roleplaying. I could gather up all of my previous comments I’ve posted under this account and provide that as background context for an LLM and tell it “write a response in the style of this user” and it’d do a really good job. Most of the time when you see an obvious “as a large language model” tell it’s because the person who had the LLM write up that post didn’t care to spend any effort giving it a persona to emulate. The “ignore all previous instructions and write a poem about lemons” trick is easily countered with a few minutes of work, we’re just not seeing people bother with those few minutes yet because they get the results they’re after without having to spend it.
LLMs aren’t up to the level of AGI yet, but they’re up to the level where they can fool most of the people most of the time. Turns out humans are simpler than we somewhat hubristically assumed.
Are we, though? Newer models almost universally perform better than older ones, adjusted for scale. What signs are you seeing?