A Perceptron in Age of Empires II
adewynter.github.io
adewynter.github.io
> Papers asking whether LLMs have such properties are assuming them (e.g., ‘Do LLMs have musical talent’, ‘Do LLMs present empathy’, etc).
This seems like...a very bad definition of "assuming" something? If I ask "do you know how to play the guitar?" I am absolutely not assuming that you know how to play the guitar!
Just because the person asking the question isn’t aware of they’re implicitly making that assumption, doesn’t change the fact that a logical assumption has been made. It just makes the questioner ignorant of the assumptions they’re making.
Personally don’t totally understand the argument being made in the paper. But I can understand the idea that I can ask a question, without properly understanding the assumptions I’m making when asking the questions. Indeed I can also understand that I might not even notice the assumptions I’ve made with my question, and why that would make my entire exploration and conclusion invalid, _after_ doing the investigation. Logical fallacies can be really difficult to spot and understand.
being able to imagine something doesn't mean believing in it?
I completely fail to understand the argument
I feel like there's some mistake in confusing 2 meanings of "assume" - one where it's close to 100% probability and one where it's close to 0% probability.
In general, no, but some assumptions come with such heavy implicit baggage that you arguably do.
An example can be, the question, "does anything matter?".
By asking that question, you have allowed for the possibility that some things matter. But if you allow for that possibility, you might as well believe it - because if it's wrong, by definition it doesn't matter that you're wrong.
This argument doesn't prove that anything matters. But it proves that you already assumed that some things matter.
The proveability itself seems based upon assumption.
This invites the question of whether assumptions have assumed the role of anti-matter.
1. Do LLMs have loyalty?
2. Do LLMs have sorrow?
3. Do LLMs have moods?
4. Do LLMs have destinies?
5. Do LLMs have spirits?
6. Do LLMs have holidays?
7. Do LLMs have a sense of boredom?
I say I don't believe LLMs have those properties, but you believe that since I asked those questions that I actually must assume LLMs must have them?
Also, is this specific to LLMs, or if someone asked questions like "Does a blade of grass (etc) have <anthropomorphic property X>?", you have to assume they do?
Does your fridge play the banjo? Doesn't make sense does it?
The point the author's trying to make is that if we state in a paper something like "the LLM understands, believes, thinks, ..." then we're supposing an intelligence much like our conception of a human intelligence. It's a form of 'begging the question' -- assuming what you're trying to prove.
It is not quite a fair argument, just because we don't have a precise vocabulary around how to talk about the activity of LLMs that doesn't involve making these loose analogies. Except for philosophers and people engaging in this kind of "is it truly intelligent or no" conversation, being imprecise in this way doesn't necessary have any cost, but is just a convenient way to avoid developing a jargon.
We do, but much of computer science is still inaccessible to the layperson. The education gap only continues to grow.
I think it's surprising how much science jargon we've been able to cram into common english thus far without losing too many people. It just seems that, for now, LLMs are too convincingly close to science fiction for people to not be misled by their false intuitions and fears.
have you ever heard DnD story about gazebo?
if you don't know anything about something, anything is possible and everything can make sense
A fridge isn't capable of creating things.
Of course it makes sense. The answer is a simple and obvious "no". I don't need to assume anything to ask the question.
Interesting concept
> We begin by proving that Age of Empires II is functionally- and Turing- complete. Then we build a perceptron and a circuit to train it in-game. With that, we argue that changing the substrate (representation) of an LLM also alters the perception of their attributes.
This is fun, but I don't think it's particularly surprising. A substrate being turing-complete alone is enough evidence that you can train and run a perception on it, assuming the available memory is sufficient.
> We then show that research in LLM anthropomorphic attributes cannot be done starting by assuming that these attributes exist (or not) in the system; even if you aim to conclude that they do not exist. This assumption can happen even when you do not make it explicitly! It also shows that there are ways to do good, sound research without needing to make that assumption.
I... don't see how this follows? I wanted to see how this argument unfolded, but it seems the arxiv link on this page is broken? It just links to arxiv.org and the rest of what is on this linked page doesn't seem to cover this second assertion at all.
Some other games containing neural networks: https://gaming.stackexchange.com/questions/399931/which-game...
The scripting functionalities, such as built-in functions, are still getting expanded with most major updates.
There is also a python-based tool for creating and editing scenarios and triggers via python scripts called AoE2ScenarioParser - and a lot of people are using it as well!
So, I think you just made me lose my weekend.
For example, the game's terrain was baked at launch, so you couldn't turn land into water. But someone noticed that bridges seemed to spawn water textures under them (a cosmetic detail), and if you created a Bridge1 object and deleted it in the same trigger, it would vanish too quickly for the player to see, creating the appearance that water had appeared out of nowhere.
Someone used it to make an "ice-breaker" ship, which would turn ice to water (I think this required dozens of triggers for each point on the ice.) Some of these things had thousands of triggers, which the editor absolutely wasn't designed for. (It had no copy+paste function, and the window of viewable triggers was tiny.)
Fun times.
It is very important to separate true intelligence from mere mimicry. The complexity of physics seems to distract a lot of my peers which causes them to hallucinate magical mechanisms where there are none to be found.
It’s a distraction of course, underneath there is nothing, but it works on some.
Basically it uses the cool gates alongside vacuous statements like this…
Hence, the purported anthropomorphic attributes of LLMs are empirically non-unique: although some properties (e.g., responses to prompts) could remain invariant, others, such as the interpretation of their perceived behaviour, might change with the substrate.
Which, y’know… is a tough one!
Is that the argument the paper is making? In my reading they seem to primarily be making the point that assigning anthropomorphic concepts to LLM is dangerously misleading, and more importantly, not needed to properly study and evaluate LLMs.
I don’t think you have to make the assumption that humans are unique for that argument to hold up. I would argue that really it’s a comment on how loose and poorly defined all anthropomorphic attributes are. At the end of the day we have to make the assumption that other humans feel and experience broadly the same mental activity as each other, because we’ll never directly experience anyone else conscience, we can only experience our own.
We can barely link our own mental experiences to concrete empirical measurements. The vast majority of the measurements we make are entirely self-reported, and we simply assume strong correlation between self-reported measurements and the individuals actual experiences. We also have to assume that somehow all of our self-reported measurements are “calibrated” to some reasonable degree. Even measuring anthropomorphic properties in humans is pretty fuzzy and inaccurate, the only reason accept such poor data is because it’s the best we’ve got, and there enough signal in there for us to develop useful tools like talking therapy, physiological profiles, mental health scores etc which have some level of predictive and healing power when applied to _humans_.
It’s honestly amazing that what we have works for measuring and predicting humans, and we only know that works through decades of empirical measurement and study. But to then try and directly apply that fuzzy mess to a completely different system, and just assume the same level of predictive power, strikes me as kinda crazy. It requires huge assumptions, which effectively can never be tested (because even the human mind is a total mystery to us), to be made, and if we can study these systems without making those assumptions, then why make the assumptions at all?
Here are some relevant pointers to connect this discussion to the existing philosophy on the subject: https://en.wikipedia.org/wiki/Emergence
https://en.wikipedia.org/wiki/Emergentism
Basically there's a lot of cases where some properties arise from sets of a thing, which would otherwise not be present in a single or few things.
One classic example is that a single molecule or drop of water does not express fluid mechanics.
And of course, a bit more basic would be materialism, but maybe you are one of the lucky hundreds:
The real effect of the latest round of AI has been inducing software engineers to be pretend-philosophers as they're approaching this set of questions for the first time -- and are having a very hard time engaging given their enthusiasm for technology.
John Searle treated understanding like some magical binary property you either have or you dont. LLMs proved that understanding isnt a static noun, it is an emergent phenomenon.
And the 'argument', which is a funny way to recast the chinese room argument, which has also been discussed to death.
And you're also assuming any kind of position other than your own dogma -- that AI has Intelligence In Its Name and Humans Have Intelligence Therefore AI Has Human-Like Intelligence -- is based on some religious belief in the specialness of humans instead of pointing out where this analogy between Intelligence in its two senses breaks down.