Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Meta Superintelligence Labs' first paper is about RAG

paddedinputs.substack.com

Post reply on HN
Print view View original

241–250 of 283 posts

Re: Meta Superintelligence Labs' first paper is about RAG

#241

Post by godelski »

what-the-grump wrote: post #232
smokel wrote: post #144

Earlier quoted context omitted.

I agree that the system itself is dysfunctional, and I understand the argument that individuals are shaped or even constrained by it. However, in this case, we are talking about people who are both exceptionally intelligent and materially secure. I think it's reasonable to expect such individuals to feel some moral responsibility to use their abilities for broader good. As for whether that expectation is "selfish" on…

I just don't think so, these exceptionally intelligent people are masters at pattern recognition, logic, hyper-focus, task completion in a field. Every single thing will tell them don't go against the flow, don't stick your neck out, don't be a hero, don't take on risk. Or you will end up nailed to a cross. To me this is an insane position to take or to expect from anyone, its some just world fallacy thing perpetuate…

  > Every single thing will tell them don't go against the flow, don't stick your neck out, don't be a hero, don't take on risk. Or you will end up nailed to a cross.
Except the situation is more like monkeys and a ladder. The ones "nailing them to the cross" are the same ones in those positions. This is the same logic as "life was tough for me, so life should be tough for you." It's idiotic!

  > So let me just stop and change the world, for what?
This is some real "fuck you, I got mine" attitude. Pulling the ladder up behind you.

We have a long history in science of seeing that sticking your neck out, taking risks, and being different are successful tools to progressing science[0]. Why? Because you can't make paradigm shifts by maintaining the current paradigm. We've also seen that this behavior is frequently combated by established players. Why? Because of the same attitude, ego.

So we've created this weird system where we tell people to think different and then punish them for doing so. Yeah, people are upset about it. I find that unsurprising. So yeah, fuck you, stop pulling the ladder up behind you. You're talking as if they just leave the ladder alone, but these are the same people who end up reviewing papers, grants, and are thus the gatekeepers of progress. Their success gives them control of the ladders and they make the rules.

[0] Galileo, Darwin, Gauss, Kepler, Einstein, and Turing are not the only members of this large club. Even more recently we have Karikó who ended up getting the 2023 Nobel prize in Medicine and Akerlof, Spence, Stiglitz who got the 2001 Nobel prize in economics for their rejected work. This seems to even be more common among Nobel laureates!

Re: Meta Superintelligence Labs' first paper is about RAG

#242

Post by godelski »

AnthonyMouse wrote: post #224
godelski wrote: post #218

Earlier quoted context omitted.

I agree with you, in a way. I've just taken another step to understand the philosophy of those bureaucrats. Clearly they have some logic, right? So we have to understand why they think they can organize and regulate from the spreadsheet. Ultimately it comes down to a belief that the measurements (or numbers) are "good enough" and that they have a good understanding of how to interpret them. Which with many bureaucrac…

In a sense you can do the same thing to yourself. If you self-impose a target and try to meet it while ignoring a lot of things that you're not measuring even though they're still important, you can unintentionally sacrifice those things. But there's a difference. In that case you have to not notice it, which sets a much lower cap on how messed up things can get. If things are really on fire then you notice right awa…

  > In a sense you can do the same thing to yourself.
Of course. I said you can do it unknowingly too.

  > The degree to which it's a problem is proportional to the size of the bureaucracy.
Now take a few steps more and answer "why". What are the reasons this happens and what are the reasons people think it is reasonable? Do you think it happens purely because people are dumb? Or smart but unintended. I think you should look back at my comment because it handles both cases.

To be clear, I'm not saying you're wrong. We're just talking about the concept at different depths.

Re: Meta Superintelligence Labs' first paper is about RAG

#243

Post by cwmoore »

throwaway314155 wrote: post #231
cwmoore wrote: post #230

Earlier quoted context omitted.

Is this a trick question? Probably before he was even born.

Is this a trick response? There's no way he ever cared about society in a way that wasn't completely plastic.

Sure, for some variation on the meaning of “society”, or “care”, or “plastic”, and maybe all the best ones, but it’s hard to argue he had never seen value in groups of people before starting Facebook, and arguably a motivator for every human being ever born.

Re: Meta Superintelligence Labs' first paper is about RAG

#244

Post by yellowcake0 »

godelski wrote: post #239
yellowcake0 wrote: post #229

Earlier quoted context omitted.

It's a pretty exotic type of addition that would lead to the second set of examples, just trying to get an idea of its nature.

Calling it addition is hairy here. Do you just mean an operator? If so, I'm with you. But normally people are expecting addition to have the full abelian group properties, which this certainly doesn't. It's not a ring because it doesn't have the multiplication structure. But it also isn't even a monoid[0] since, as we just discussed, it doesn't have associativity nor unitality. There is far less structure here than y…

I think you misinterpreted the tone of my original comment as some sort of gotcha. Presumably you're overloading the addition symbol with some other operational meaning in the context of vector embeddings. I'm just calling it addition because you're using a plus sign and I don't know what else to call it, I wasn't referring to addition as it's commonly understood which is clearly associative.

Re: Meta Superintelligence Labs' first paper is about RAG

#245

Post by dang »

lanyard-textile wrote: post #234
dang wrote: post #227

Earlier quoted context omitted.

There is of course such a guideline: https://news.ycombinator.com/newsguidelines.html We don't catch every case, but if you're talking about the frontpage, I'm surprised to hear you say "epidemic". What are some recent examples?

I wouldn’t give much weight to the person that had an opinion about the guidelines without reading them :)

Acknowledging their existence in principle is already a lot!

Re: Meta Superintelligence Labs' first paper is about RAG

#246

Post by danielmarkbruce »

smokel wrote: post #134
crystal_revenge wrote: post #80

Earlier quoted context omitted.

I've spent most of my career working, chatting and hanging out with what might be best described as "passionate weirdos" in various quantitative areas of research. I say "weirdos" because they're people driven by an obsession with a topic, but don't always fit the mold by having the ideal combination of background, credentials and personality to land them on a big tech company research team. The other day I was spend…

> I certainly didn't judge them because they are just playing the game. Please do judge them for being parasitical. They might seem successful by certain measures, like the amount of money they make, but I for one simply dislike it when people only think about themselves. As a society, we should be more cautious about narcissism and similar behaviors. Also, in the long run, this kind of behaviour makes them an annoyi…

There is an implication that passionate weirdos are good by nature. You either add value in the world or you don't. A passionate, strange actor or musician who continues trying to "make it" who isn't good enough to be entertaining is a parasite and/or narcissist. A plumber who is doing the job purely for money is a value add (assuming they aren't ripping people off) - and they are playing the game - the money for work game.

Re: Meta Superintelligence Labs' first paper is about RAG

#247

Post by danielmarkbruce »

yellowcake0 wrote: post #244
godelski wrote: post #239

Earlier quoted context omitted.

Calling it addition is hairy here. Do you just mean an operator? If so, I'm with you. But normally people are expecting addition to have the full abelian group properties, which this certainly doesn't. It's not a ring because it doesn't have the multiplication structure. But it also isn't even a monoid[0] since, as we just discussed, it doesn't have associativity nor unitality. There is far less structure here than y…

I think you misinterpreted the tone of my original comment as some sort of gotcha. Presumably you're overloading the addition symbol with some other operational meaning in the context of vector embeddings. I'm just calling it addition because you're using a plus sign and I don't know what else to call it, I wasn't referring to addition as it's commonly understood which is clearly associative.

You guys are debating this as though embedding models and/or layers work the same way. They don't.

Vector addition is absolutely associative. The question is more "does it magically line up with what sounds correct in a semantic sense?".

Re: Meta Superintelligence Labs' first paper is about RAG

#248

Post by edanm »

schmorptron wrote: post #120

One thing I don't get about the ever-reoccuring RAG discussions and hype men proclaiming "Rag is dead", is that people seem to be talking about wholly different things? My mental model is that what is called RAG can either be: - a predefined document store / document chunk store where every chunk gets a a vector embedding, and a lookup decides what gets pulled into context as to not have to pull whole classes of docu…

> My mental model is that what is called RAG can either be:

RAG is confusing, because if you look at the words making up the acronym RAG, it seems like it could be either of the things you mentioned. But it originally referred to a specific technique of embeddings + vector search - this was the way it was used in the ML article that defined the term, and this is the way most people in the industry actually use the term.\

It annoys me, because I think it should refer to all techniques of augmenting, but in practice it's often not used that way.

There are reasons that specifically make the "embeddings" idea special - namely, it's a relatively new technique that actually fits LLM very well, because it's a semantic search - meaning, it works on "the same input" as LLMs do, which is a free-text query. (As opposed to a traditional lookups that work on keyword search or similar.)

As for whether RAG is dead - if you mean specifically vector-embeddings and semantic search, it's possible - because you could theoretically use other techniques for augmentation, e.g. an agent that understands a user question about a codebase and uses grep/find/etc to look for the information, or composes a search to search the internet for something. But it's definitely not going to die in that second sense of "we need some way to augment LLMs knowledge before text generation", that will probably always be relevant, as you say.

Re: Meta Superintelligence Labs' first paper is about RAG

#249

Post by yellowcake0 »

danielmarkbruce wrote: post #247
yellowcake0 wrote: post #244

Earlier quoted context omitted.

I think you misinterpreted the tone of my original comment as some sort of gotcha. Presumably you're overloading the addition symbol with some other operational meaning in the context of vector embeddings. I'm just calling it addition because you're using a plus sign and I don't know what else to call it, I wasn't referring to addition as it's commonly understood which is clearly associative.

You guys are debating this as though embedding models and/or layers work the same way. They don't. Vector addition is absolutely associative. The question is more "does it magically line up with what sounds correct in a semantic sense?".

I'm just trying to get an idea of what the operation is such that man - man + man = woman, but it's like pulling teeth.

Re: Meta Superintelligence Labs' first paper is about RAG

#250

Post by AnthonyMouse »

godelski wrote: post #242
AnthonyMouse wrote: post #224

Earlier quoted context omitted.

In a sense you can do the same thing to yourself. If you self-impose a target and try to meet it while ignoring a lot of things that you're not measuring even though they're still important, you can unintentionally sacrifice those things. But there's a difference. In that case you have to not notice it, which sets a much lower cap on how messed up things can get. If things are really on fire then you notice right awa…

> In a sense you can do the same thing to yourself. Of course. I said you can do it unknowingly too. > The degree to which it's a problem is proportional to the size of the bureaucracy. Now take a few steps more and answer "why". What are the reasons this happens and what are the reasons people think it is reasonable? Do you think it happens purely because people are dumb? Or smart but unintended. I think you should…

I don't think the premise that everything is a proxy is right. We can distinguish between proxies and components.

A proxy is something like, you're trying to tell if hiring discrimination is happening or to minimize it so you look at the proportion of each race in some occupation compared to their proportion of the general population. That's only a proxy because there could be reasons other than hiring discrimination for a disparity.

A component is something like, a spaceship needs to go fast. That's not the only thing it needs to do, but space is really big so going fast is kind of a sine qua non of making a spaceship useful and that's the direct requirement rather than a proxy for it.

Goodhart's law can apply to both. The problem with proxies is they're misaligned. The problem with components is they're incomplete. But this is where we come back to the principal-agent problem.

If you could enumerate all of the components and target them all then you'd have a way out of Goodhart's law. Of course, you can't because there are too many of them. But, many of the components -- especially the ones people take for granted and fail to list -- are satisfied by default or with minimal effort. And then enumerating the others, the ones that are both important and hard to satisfy, gets you what you're after in practice.

As long as the person setting the target and the person meeting it are the same person. When they're not, the person setting the target can't take anything for granted because otherwise the person meeting the target can take advantage of that.

> What are the reasons this happens and what are the reasons people think it is reasonable? Do you think it happens purely because people are dumb? Or smart but unintended.

In many cases it's because there are people (regulators, corporate bureaucrats) who aren't in a position to do something without causing significant collateral damage because they only have access to weak proxies, and then they cause the collateral damage because we required them to do it regardless, when we shouldn't have been trying to get them to do something they're in no position to do well.

Post reply on HN