site banner
PaperclipPerfector  19d ago (text post)   32922 thread views

Culture War Roundup for the week of November 4, 2024

This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.

Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.

We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:

  • Shaming.

  • Attempting to 'build consensus' or enforce ideological conformity.

  • Making sweeping generalizations to vilify a group you dislike.

  • Recruiting for a cause.

  • Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.

In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:

  • Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.

  • Be as precise and charitable as you can. Don't paraphrase unflatteringly.

  • Don't imply that someone said something they did not say, even if you think it follows from what they said.

  • Write like everyone is reading and you want them to be included in the discussion.

On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.

8
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

Speculation: It’s interesting that the bottleneck is given as lack of data rather than architecture. That opens up the possibility that we may be able to get things moving again by finding some other method of obtaining/creating useable data.

LLMs were historically created to use next-token-prediction as a means of solving natural language processing tasks. I think we can regard that problem as provisionally solved. When people talk about GPTs limits, they aren’t talking about its ability to take English input and produce readable English output. They are talking about general intelligence: the ability to output sensible, useful English output.

In short, LLMs are general learning machines using natural language as a proxy task. Natural language is cheap and information rich but any means of conveying information about the world is fair game, provided that it can be converted into the same token space that GPT is using using CLIP or something similar.

What is needed is large quantities of data that conveys causal information about the world. Video is probably a good place to start. Some kind of simulated self-play might also be useable. What else could be useable?

(I’m not sure how next-token prediction would work here)

It’s interesting that the bottleneck is given as lack of data rather than architecture.

It's not lack of data as such (there's gobs and gobs of raw data). It's curated high quality data of a suitable form and that can actually be used (be that for legal or technical reasons). The reason synthetic data is used because it solves (or claims to solve) the curation and form issues. The trainers can directly instruct the source AI to provide data of type X in quantity Y.

Indeed, but this introduces its own problems. This is arguably a large part of why Google's AI products are noticeably more prone to "hallucination" than thier immediate peers.

Of course. I'd estimate that using synthetic data results in an overall worse performing AI but I could see it being used to fill specific gaps in the real training material (probably using a specialized model that's good at that specific type of data and possibly not much else).

What else could be useable?

Somewhat speculative, but non-invasive recording of brain activity seems like a promising underutilized modality. When sufficiently discreet devices reach the market -- say, for controlling your phone -- they would be worn anyway, continuously throughout the day, so just add a few more lines about personal data collection in a license agreement. To get labeled data, make an app which prompts humans with various signals and records their reaction. Gamify, pay if needed, etc. Seems scalable.

In effect LLMS aren't smart, they are just great at recognizing patterns they are trained on. Google is great at recognizing text strings that it remembers, LLMS don't need matching strings they match on patterns and are able to combine patterns from multiple sources. LLMs aren't truly intelligent because they are dumbfounded if there isn't a good matching pattern in the training set. They are stumped in a way a human isn't if they encounter something new.

LLMs aren't going to replace humans because the set of all data is miniscule to the set of all potential patterns in the world.

LLMs aren't going to replace humans because the set of all data is miniscule to the set of all potential patterns in the world.

I mean, you can say LLMs aren't going to replace humans...but the 'potential patterns in the world' are all reducible to data in one way or another.

So some Machine trained on language AND physics data AND biology AND etc. etc. is still a potential contender, no?

the 'potential patterns in the world' are all reducible to data in one way or another

I mean is it? Quantitative Realism doesn't exactly seem self evident.

I've consistently pointed AI hype believers to their own metaphysical assumptions and this is the crux of it.

Are we just pattern matching engines or does agency have another source and is that in anyway connected to our experience of consciousness?

I think when people believed that larger gizmoes we don't fully understand would give us the answer to this question, they were deluding themselves, and I'm somewhat dissapointed that I was right since we are still without answers. But at least the possibility that we have a soul, ghost or another manner of special thing that automata don't is still secure.

Now the real test will be this: if Musk can convince enough people to use Neuralink and get their brain patterns recorded 24/7, and if someone trains transformers on that, what will be the outcome? Can we Chinese room our way to general intelligence?

I don't know, but it seems like the most logical way forward, since access to immense unpolluted datasets is no longer a possibility.

I mean is it? Quantitative Realism doesn't exactly seem self evident.

Isn't Computational Complexity Theory supposed to tackle questions of this kind?

Scott Aaronson offered the following highly evocative metaphor:

The best definition of complexity theory I can think of is that it’s quantitative theology: the mathematical study of hypothetical superintelligent beings such as gods. Its concerns include:

  • If a God or gods existed, how could they reveal themselves to mortals? (IP=PSPACE, or MIP=NEXP in the polytheistic case.)
  • Which gods are mightier than which other gods? (PNP vs. PP, SZK vs. QMA, BQPNP vs. NPBQP, etc. etc.)
  • Could a munificent God choose to bestow His omniscience on a mortal? (EXP vs. P/poly.)
  • Can oracles be trusted? (Can oracles be trusted?)

And of course:

  • Could mortals ever become godlike themselves? (P vs. NP, BQP vs. NP.)

Although I doubt such general questions and theories are that helpful in guiding our research: they provide boundaries for what is possible, but what is practical typically lies far away from those boundaries.

Scott's metaphor is funny to think about but it has no philosophical rigor.

Complexity theory is not meaningfully different from other mathematics in its relationship to the metaphysical: it's a pure reason construct that attempts to map out necessary truths.

In many ways it it actually completely disconnected from the question at hand, because the machines it is concerned about are abstractions that are not and cannot possibly be real. They just happen to map onto real objects in a useful enough way. As you point out.

Scott isn't the first to connect this type of endeavor and the sacred. Pythagoras did it a long time ago. But the connection isn't relevant to the question of intelligence in my view.

I think pure data is a directionally useful way of looking at the world, and useful for most problem-solving purposes. I am a theist so I think there’s more beyond just physical reality, but whether or not it’s true, I think that for most projects, reducing the universe to data is going to work just fine. Consciousness is produced in the brain, and definitely experienced there, so I think you can get something like a conscious AI simply by recreating a brain. Might be easier to start with a dog or something like that, but I think even though there’s a metaphysical aspect to consciousness, that doesn’t mean that there’s no point to studying it in brains.

I suppose it comes down to whether or not there is a ghost in the machine.

If human intelligence is all neurons that can be modeled as a graph with weighted edges then we should be able to simulate it.

Maybe we do that and still can’t get human intelligence to pop out of the simulated brain and find that something is missing.

It would be a bit funny if they design a machine that is provably a 1:1 simulation of a human brain, switch it on, and get an error message to the effect of "Cannot Execute Commands: This unit is not ensouled."

“Humunculus not installed: please refer to manual.”

I mean, that kind of sounds like you're saying it's provably not a 1:1 simulation of a human brain.

What you're describing is measurable evidence of new physics. Every physicist in the world would want to buy you a beer.

...Please contact your local soul provider for further information

If you believe that you've received this message in error, please contact your system administrator at t0.yahweh.root.

Control thread not found.

Finally, a reason for MMOGs to come back to the mainstream as data production interfaces.

That EvE online cell structure minigame was ahead of its time.

We’re exhausting almost all the data, video included. We’ve recently taken to generating synthetic data. For images, this would mean generating novel images and then feeding them back into training. Imagine taking an image of a car and then rotating it behind some thick leaves or a chain link fence.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats