51–60 of 104 posts
I don’t know how well this works (demo is broken on mobile), but I like the idea. Imagine an infinite wiki where articles are generated on the fly (from reputable sources - with links), including links to other articles (which are also generated) etc. I actually like this sort of interface more than chat.
I saved a full snapshot of Wikipedia (and Stack Overflow) in the weeks before ChatGPT launched, and every day I'm more glad that I did. They will become the Low Background Steel of text.
That's gonna be a lot of fun to play with in a year or so. There's a concurrent explosion of 'veracity' analysis - it'll be fun to run those against wikipedia a year from now and your data. Incidentally, are you interested in mirroring your dataset and making it more robust? I'm sure I've got a few TB of storage lying around somewhere...
Earlier quoted context omitted.
There's an important difference between wikipedia and the LLMs that are actually useful today. Wikipedia is open, like completely open. GPT is not. Unless we manage to crack the distributed training / incremental improvement barriers, LLMs are a lot more likely to follow the Google path (that is, start awesome and gradually enshittify as capitalist concerns pollute the decision matrix) than they are the Wikipedia pat…
this is super interesting! it also carves I to the question what constituted model openness? most people agree that just releasing weights are not enough. but I don't think it will ever be feasible to say that reproducing model training is feasible. especially when factoring in branching and merging of models. for me this is an open and super interesting question.
A model that can be incrementally trained (this is the bit we're missing) hosted by a nonprofit, belonging to "we the people" (like wikipedia).
The training process could be done a little like wikipedia talk pages are now - datasets are proposed and discussed out in the open and once generally approved, trained into the model.
Because training currently involves backpropagation, this isn't possible. Hinton was working on a structure called "forward-forward" that would have overcome this (if it worked) before he decided humanity couldn't be trusted [1]. It is my hope that someone smarter than me picks up this thread of research - although in the spirit of personal responsibility I've started picking up my old math books to try and get to a point where I grok the implementation enough to experiment myself (I'm not super confident I'm gonna get there but you can't win if you don't play, right?)
It's hard to tell when (if?) we're ever going to have this - if it does happen, it'll be because a lot of people do a lot of really smart unpaid work (after seeing OpenAI do what it did, I don't have a ton of faith that even non-profit orgs have the will or the structure to pull it off. Please prove me wrong.)
Earlier quoted context omitted.
The concern is not just a vaguely cynical hand-wringing about how bad AI is. Feeding AIs their own output as training material is a bad thing for mathematical reasons, and feeding AIs the output of other very similar AIs is close enough for it to also be bad. The reasons are subtle and hard to describe in plain English, and I'm not enough of an expert to even try, so pardon if I don't. But given that it is hard to de…
>Feeding AIs their own output as training material is a bad thing for mathematical reasons Most model collapse studies explore degenerate cases to determine the potential limits of the training process of the same model. No wonder you will get terrible results if you recursively recompress a JPEG 100 times! In real world it's nowhere near that bad, because models are never trained on their output alone and always gua…
However, if AIs become as successful as Nvidia stock price implies, it could indeed become difficult to find text that is guaranteed to not be AI. It is conceivable that in 20 years it will be very difficult to generate a training set at any scale that isn't 90% already touched by AIs.
Of course, it's conceivable that in 20 years we'll have AIs that don't need the equivalent of millennia of training to come up to their full potential. The problem is much more tractable if one merely needs to produce megabytes of training data to obtain a decent understanding of English rather than many gigabytes.
I looked into this to see where it was getting new information, and as far as I can tell, it is searching wikipedia exclusively. Useful for sure, but not exactly what I was expecting based on the title.
For broader implementation you would want to develop the approach further. The idea of sampling other-language Wikipedia mentioned in a sibling comment seems to be a decent next step.
Extending it to bringing in from wider sources would be another step. I doubt it would be infallible but it would be really interesting to see how it compares to humans performing the same task. Especially if there were a additional ability to verify written articles and make corrections.
Earlier quoted context omitted.
On the one hand, a tool is as good or bad as the person wielding it. Smart folks with the right intentions will certainly be able to use this stuff to increase the rate and quality of their output (because they're smart, so they'll verify rather than trust. Hopefully.) On the other, moderation is an unsolved problem. The general mess of the internet is probably not quite ready to be handed a footgun of this caliber.…
> On the one hand, a tool is as good or bad as the person wielding it. I think the real reason is one line dogmas like this.
To be clear - I'm with you that these systems can absolutely be a force for vast good (at least, I think that was what you were getting at unless there was a missing '/s'). I use them daily to pretty astounding effect.
I'll admit to being a little put off by being labeled dogmatic - it's not something I consider myself to be.
I looked into this to see where it was getting new information, and as far as I can tell, it is searching wikipedia exclusively. Useful for sure, but not exactly what I was expecting based on the title.
As a base for researching the idea, Wikipedia seems like a decent data source. For broader implementation you would want to develop the approach further. The idea of sampling other-language Wikipedia mentioned in a sibling comment seems to be a decent next step. Extending it to bringing in from wider sources would be another step. I doubt it would be infallible but it would be really interesting to see how it compare…
If your goal is to generate a wiki article, you can't assume one already exists. That's begging the question. If you could just search wikipedia for the answer, you wouldn't need to generate an article.
I can see this being useful iif the content is generated on demand and then discarded. Publishing AI generated material is generally speaking a horrible idea and does nobody any good (at least until accuracy levels get much much better.) Even if they do it well and truthfully (which they don't) current LLMs can only summarize, digest, and restate. There is no non-transient value add. LLMs may have a place to help que…
Though, at a stretch, Wikipedia itself could be considered based around summarization, digesting, and restating/citing things said elsewhere, given its policy of verifiability: "Even if you are sure something is true, it must have been previously published in a reliable source before you can add it." Now, LLMs aren't well known for their citation skills, to be fair.. :-)
Earlier quoted context omitted.
can you show me a mathematical reason that cannot philosophically be applied to people also? people only being fed other people output.
I'd go with "no", because people just consuming the output of other people is a big ongoing problem. Input from the universe needs to be added in order to maintain alignment with the universe, for whichever "universe" you are considering. Without frequent reference to reality, people feeding too much on people will inevitably depart from reality. In another context, you may know this as an "echo chamber". Not quite e…
currently we have mostly LLMs in the mix. but there are no reason that the Ai mix will not contain embodied agents thst also publish stuff in the internet. (think search and rescue bots that automatically write a report).
Now Ai is connected to reality without people in the mix.
Earlier quoted context omitted.
As a base for researching the idea, Wikipedia seems like a decent data source. For broader implementation you would want to develop the approach further. The idea of sampling other-language Wikipedia mentioned in a sibling comment seems to be a decent next step. Extending it to bringing in from wider sources would be another step. I doubt it would be infallible but it would be really interesting to see how it compare…
> As a base for researching the idea, Wikipedia seems like a decent data source. If your goal is to generate a wiki article, you can't assume one already exists. That's begging the question. If you could just search wikipedia for the answer, you wouldn't need to generate an article.