> LLMs are not particularly good at arithmetic, counting syllables, or recognizing haikus, though, because (contrary to the thesis of the article) they don’t magically acquire whatever ability would “simplify” predicting the next token.
LLMs understand it to a certain extent. It's more then "predicting" the next token. When people ascribe "predicting the next token" it's a niave and unintelligent description to cover up what they don't understand.
I mean you can describe a human brain as simply wetware, a jumble of signals and chemical reactions that twitch muscles and react to pressure waves in the air and light. But obviously there is a higher level description of the human brain that is missing from that description.
The same thing could be said about LLMs. I can tell you this, researchers completely understand token prediction that much can be said. What we don't currently understand is the high level description. Perhaps it's not something we can understand as we've never been able to understand human consciousness at a high level either.
That's the thing with people. Nobody actually understands the high level description of a fully trained LLM. People are lambasting others because they "think" they understand when they only actually understand the low level primitives. We understand assembly, but you don't understand the Operating system written in assembly.
The chances of chatGPT memorizing or even predicting the next tokens here are in a probability too low to even consider. There are so many possible numbers here even numbers that aren't true but have a "higher probability" of being close to the truth from a token/edit-distance standpoint. It's safe to say, from a scientific standpoint, chatGPT in this scenario understands what it means to add 1.
Realize that this calculation results in an overflow. chatGPT needs symbolic understanding to perform the feat it did above.
But there are, of course, things it gets wrong. But again we don't truly understand what's going on here. Is it lying to us? Perhaps it can't differentiate between just a generated statistical token or a actual math equation. It's hard to say. But from the example above, by probability, we know that an aspect of true understanding and ability exists.
I do think that LLMs have emergent properties that do some interesting things, however I would like to point out simple next token prediction would work on your example quite well.
<numbers>0 + 1 -> <numbers>1
Even simple attention mechanisms would handle that quite well with enough examples of <numbers>
4320598340958340958340953095809348509348503480958340958304985038530999999 + 1 ?
The sum of 4320598340958340958340953095809348509348503480958340958304985038530999999 and 1 is 4320598340958340958340953095809348509348503480958340958304985038531000000.
which is more complex.
I'm too lazy to get it to add two large numbers together.
Also, I've never been convinced that "ability to do arithmetic" has any relationship to intelligence. We don't expect regular humans to be able to add two large numbers together reliably, either.
This number is tokenized as a list: 43 20 59 83 409 58 340 9 58 340 95 30 95 809 34 850 9 34 850 34 809 58 340 9 58 30 49 850 385 30 49 58 30. If GPT recognizes the context "+ 1 equals" through the attention mechanism, it can predict that the next number in the sequence should be 31: ... 58 30 -> ... 58 31
This is good for procedural generated 2D worlds. Think Hollow Knight, but expansive across infinite environments. Just randomly generate the control image and have the LLM generate the theme. Combine that with LLM generated lore and the possibilities are unlimited.
I have far more simpler (I imagine?) case already in mind, from the recent Cities discussion thread:
>> I would've expected at least a not grid-based zoning so that buildings on curves look more natural. All these empty pieces of land in between buildings look really bad and kind of force us to make grid cities. And that is not even an innovation, it was already present in the SimCity series. But some procedurally generated buildings for smooth corners and connecting buildings would be nice.
> Its hard to make assets that would work with every curve. When you see screenshots of nice cities like this, people are using mods to hand place assets with them clipping into each other to make a unified wall of buildings along the curve or corner.
ML generated building configurations for city builder games. Readily adaptable to any shape, and as a bonus can break up excessive repetition a bit. If you want to be ambitious, train a model on real-world aerial photos.
Would this require ML? I would think this could be accomplished with some rudimentary procedural generation. Divide the large space up into building sized plots, then generate a building to fit the space.
Yeah, in the map editor there is, in fact a random button that generates. I havn't gotten around to making sure that the random level is playable (and about 1 in 4 have unreachable areas) but that wouldn't be that hard to add. (I've been focused on the creative aspect of creating your own levels because right now that part is more fun).
There's no books man. This stuff is too new. But all the components are in place and exist.
This guy just demonstrated what's required to generate a theme, and it's not far off from extrapolating further from that in using a LLM to generate Lore and just some random maze generator to create the base control image.
Doubtful. The developer is really old school. Additionally he's already spent years on integrating his huge lore generator into the fabric of the game. An LLM will throw a wrench into the whole process.
Loma Linda is a bluezone because of diet and community. They are very far away from the ocean. The environment their is more desert rather then Mediterranean climate.
How do I know? I live in LA county about a couple miles due west of loma linda. That city is really inland and not considered to be near the ocean at all.
> The pyrex (all lowercase) trademark is now used for kitchenware sold in the United States, South America, and Asia. In Europe, Africa, and the Middle East, a variation of the PYREX (all uppercase) trademark is licensed by International Cookware for bakeware that has been made of numerous materials including borosilicate and soda-lime glass, stoneware, metal, plus vitroceramic cookware.
>Much less commonly said, but still true, is that extraordinary claims require extraordinary trust.
What you say is true. But here's the problem. 99% of people in this world are people you don't know well and don't know personally and they aren't famous enough for you to trust them.
Thus by probability an extraordinary claim that is true is highly, highly likely to come from someone you will never have any form of "extraordinary" trust for. So your statement, which I quoted, is, from a practical standpoint, completely useless. By your logic all extraordinary claims in existence will all pretty much be mistrusted by you. It's a statement of complete denial of anything that sounds outlandish. It's the definition of being closed minded.
Being extremely closed minded tends to be better than being extremely open minded where you believe every freaking thing you hear. But the best place to be in is at neither extreme. Rather then deny all extraordinary claims you should be skeptical.
You need to search for techniques in which will help you determine whether you CAN or CANNOT trust someone you do not already have "extraordinary" trust for.
To which I throw this statement at you:
If someone makes a statement without any ulterior motive that results in net benefit to them, you can place more merit on the fact that they are telling you what they believe to be the truth.
So take for example a stranger tells you there's a bomb in the building and you have to get out. You get the hell out because that stranger had nothing to gain from telling there's a bomb.
If a person with a camera is recording you and he told you there's a bomb in the building. Well that's different right? The key difference here is that in one situation there's a clear ulterior motive. In the other situation there was no motive.
Follow the motive.
The person making the claim... what is his motive?
> ...and they aren't famous enough for you to trust them.
That's not how we gain extraordinary trust (remember, Kyrie Irving is pretty darn famous).
We expect all statements to be independently verified and attested by a bunch of experts, with eg. journalists verifying they are indeed experts (that's journalists' expertise). The more people who've you learned to trust in that chain of verifiers, the higher your trust.
When it comes to extraordinary claims like this, I usually go back to the "implementation": why and how did another civilization solve all of the physics problems of interstellar flight (if it did), how did they go undetected by other countries and amateurs, and why did they not simply reach out in an attempt to communicate?
At that point I can even buy that these "intelligence officers" believe what they are saying, but I still don't buy that their claims are true.
>That's not how we gain extraordinary trust (remember, Kyrie Irving is pretty darn famous).
Nope. It is how you gain extraordinary trust. It really depends on what they are famous for. There are many famous people or scholars who you respect and trust due to their fame. You trust them, because others trust them.
Fame can also cause distrust too. But it does cause trust as well.
>When it comes to extraordinary claims like this, I usually go back to the "implementation": why and how did another civilization solve all of the physics problems of interstellar flight (if it did), how did they go undetected by other countries and amateurs, and why did they not simply reach out in an attempt to communicate?
This is too much speculation. There are an infinite number of explanations here. Better to follow the the chain of events from the anomalous claim and start piecing together things forwards rather then just claim it's impossible because you can't think of any valid way for it to happen.
Additionally the claim itself is "extraordinary" so basically there is likely no mundane explanation for it if the claim is true.
>At that point I can even buy that these "intelligence officers" believe what they are saying, but I still don't buy that their claims are true.
So that's the first step right? They believe what they are saying. So from their perspective they witnessed extraordinary evidence. So what made them believe it? What did they actually see? Are they correct? Even if following this thread leads to something different then their claim, one can still learn something interesting.
Closed mindedness often involves logic that works backwards. You need to move forwards. If you see a path, moving forwards means following the path. Moving backwards means ignoring the path because you made a guess that there's nothing at the end of the path.
I am not sure why do you think there is only one way to increase trust in a claim made by unknown people?
I don't trust "famous" people. I trust that it's unlikely for a large group of "famous" (eg experts in relevant fields, which is a more specific interpretatiin of "famous"), unrelated people to not trust each other, and only once they reach a common ground, would my trust of their extraordinary claim rise.
Yes, I wouldn't trust Einstein proclaiming speed of light is finite, but when corroborated with independent claims of experiments from Michelson proving likewise, and others from 30 years ago, I would.
Basically, if it's too few people, I'd question how did others, unrelated people not see it: science mostly progresses these days when the body of knowledge is such that simply the next thing is in front of us, and multiple people simultaneously "discover" it. As it's many people making similar claims, I'd start considering it seriously.
This is how science works, and I am happy to follow that with media reports as well. It's not at all backwards IMO.
Sure, if I was working on a hypothesis, I would certainly venture into less trusty sources, but I would need to be ready to accept for my hypothesis to be disproven too.
You ignored the fact these people are not neutral. They're liars. If they're telling the truth now, they're some of the largest liars in human history. Your hypothetical is irrelevant in this case.
I have answers, but discussing them would just be further excuses to try to dilute away this fundamental truth, so I decline. It's their responsibility to figure out how to be more trustworthy if they want it, not mine. (Helpful tip: They can start by not lying about everything, all the time.) It's my responsibility not to trust known, repeated, huge liars.
LLMs understand it to a certain extent. It's more then "predicting" the next token. When people ascribe "predicting the next token" it's a niave and unintelligent description to cover up what they don't understand.
I mean you can describe a human brain as simply wetware, a jumble of signals and chemical reactions that twitch muscles and react to pressure waves in the air and light. But obviously there is a higher level description of the human brain that is missing from that description.
The same thing could be said about LLMs. I can tell you this, researchers completely understand token prediction that much can be said. What we don't currently understand is the high level description. Perhaps it's not something we can understand as we've never been able to understand human consciousness at a high level either.
That's the thing with people. Nobody actually understands the high level description of a fully trained LLM. People are lambasting others because they "think" they understand when they only actually understand the low level primitives. We understand assembly, but you don't understand the Operating system written in assembly.
Take this for example:
The chances of chatGPT memorizing or even predicting the next tokens here are in a probability too low to even consider. There are so many possible numbers here even numbers that aren't true but have a "higher probability" of being close to the truth from a token/edit-distance standpoint. It's safe to say, from a scientific standpoint, chatGPT in this scenario understands what it means to add 1.Realize that this calculation results in an overflow. chatGPT needs symbolic understanding to perform the feat it did above.
But there are, of course, things it gets wrong. But again we don't truly understand what's going on here. Is it lying to us? Perhaps it can't differentiate between just a generated statistical token or a actual math equation. It's hard to say. But from the example above, by probability, we know that an aspect of true understanding and ability exists.