gray
@fusha.moe
I push the pretty pixels 🖌️✨ I move my body in fun ways ⛸️⛷️🤸♀️🏍️ I do rendering, engine stuff, and tech art with Rust at Embark Studios 🦀👩💻
Reposted by gray
i was also struck by how many people receiving SNAP are employees of wal-mart and other big corporations. almost seems like we taxpayers are subsidizing private companies 🤔
November 9, 2025 at 3:14 AM
i was also struck by how many people receiving SNAP are employees of wal-mart and other big corporations. almost seems like we taxpayers are subsidizing private companies 🤔
Halo really was goated in this regard (and so many others)
August 7, 2025 at 9:02 AM
Halo really was goated in this regard (and so many others)
I just love your art soooo much. Thank you for sharing it with all of us
July 24, 2025 at 8:54 PM
I just love your art soooo much. Thank you for sharing it with all of us
In some sense of the phrase, yes I agree. But I think a simulation of the actual experience of human emotions is distinct from simulating a plausible way for emotions to be expressed, and I don't think we can assume LLMs are doing the former when we observe them doing the latter.
July 24, 2025 at 9:42 AM
In some sense of the phrase, yes I agree. But I think a simulation of the actual experience of human emotions is distinct from simulating a plausible way for emotions to be expressed, and I don't think we can assume LLMs are doing the former when we observe them doing the latter.
to be similarities, but I think *expression of* {emotion, intelligence, logic} through language is a markedly different thing than the simulation of the actual experiences/mechanisms behind those things in the way they exist in humans or what we'd otherwise think of as intelligent things.
July 24, 2025 at 9:38 AM
to be similarities, but I think *expression of* {emotion, intelligence, logic} through language is a markedly different thing than the simulation of the actual experiences/mechanisms behind those things in the way they exist in humans or what we'd otherwise think of as intelligent things.
approximates the appearance of emotions in English text"
is actually quite large and diverse, and that in particular it is quite distinct from
"the function which simulates the actual underlying emotional and rational thinking mechanisms of intelligent creatures"
Certainly I would expect there
is actually quite large and diverse, and that in particular it is quite distinct from
"the function which simulates the actual underlying emotional and rational thinking mechanisms of intelligent creatures"
Certainly I would expect there
July 24, 2025 at 9:38 AM
approximates the appearance of emotions in English text"
is actually quite large and diverse, and that in particular it is quite distinct from
"the function which simulates the actual underlying emotional and rational thinking mechanisms of intelligent creatures"
Certainly I would expect there
is actually quite large and diverse, and that in particular it is quite distinct from
"the function which simulates the actual underlying emotional and rational thinking mechanisms of intelligent creatures"
Certainly I would expect there
Thanks for the reply! I do agree with a lot of this nuance. In some sense we can definitely confidently say there is simulation going on, and from a technically correct perspective I think it's also fine to say a simulation of emotions.
But! I would posit that "set of functions whose output
But! I would posit that "set of functions whose output
July 24, 2025 at 9:38 AM
Thanks for the reply! I do agree with a lot of this nuance. In some sense we can definitely confidently say there is simulation going on, and from a technically correct perspective I think it's also fine to say a simulation of emotions.
But! I would posit that "set of functions whose output
But! I would posit that "set of functions whose output
or "having emotions," and this is the thing the "just statistics" folks are badly trying to emphasize.
It is all statistics, but there's no "just" about it. The fact that statistics can create such emergent behavior is indeed incredibly interesting and powerful and so shouldn't be taken lightly.
It is all statistics, but there's no "just" about it. The fact that statistics can create such emergent behavior is indeed incredibly interesting and powerful and so shouldn't be taken lightly.
July 23, 2025 at 9:22 PM
or "having emotions," and this is the thing the "just statistics" folks are badly trying to emphasize.
It is all statistics, but there's no "just" about it. The fact that statistics can create such emergent behavior is indeed incredibly interesting and powerful and so shouldn't be taken lightly.
It is all statistics, but there's no "just" about it. The fact that statistics can create such emergent behavior is indeed incredibly interesting and powerful and so shouldn't be taken lightly.
fake glitched and gaslighting are known and "intended" features of the models for the reasons you describe! But people should also be very clear it's all in *appearance of* all these behaviors brought about in an extremely different process than anything else we would think of as "intelligent"
July 23, 2025 at 9:22 PM
fake glitched and gaslighting are known and "intended" features of the models for the reasons you describe! But people should also be very clear it's all in *appearance of* all these behaviors brought about in an extremely different process than anything else we would think of as "intelligent"
"simulation" of emotions, of conscious thought like a living thing would etc.
We only have *output which looks like* the output that a conscious thing that *was* simulating emotions would look like.
And indeed it is important people know that the *appearance of* intelligence, emotion,
We only have *output which looks like* the output that a conscious thing that *was* simulating emotions would look like.
And indeed it is important people know that the *appearance of* intelligence, emotion,
July 23, 2025 at 9:22 PM
"simulation" of emotions, of conscious thought like a living thing would etc.
We only have *output which looks like* the output that a conscious thing that *was* simulating emotions would look like.
And indeed it is important people know that the *appearance of* intelligence, emotion,
We only have *output which looks like* the output that a conscious thing that *was* simulating emotions would look like.
And indeed it is important people know that the *appearance of* intelligence, emotion,
misinterpretation by the average person.
As far as we know, while the models capabilities *are* extremely interesting and as you say the ability to write extremely convincing language about pretty much anything is in itself very powerful, we have no evidence that there is actually even any
As far as we know, while the models capabilities *are* extremely interesting and as you say the ability to write extremely convincing language about pretty much anything is in itself very powerful, we have no evidence that there is actually even any
July 23, 2025 at 9:22 PM
misinterpretation by the average person.
As far as we know, while the models capabilities *are* extremely interesting and as you say the ability to write extremely convincing language about pretty much anything is in itself very powerful, we have no evidence that there is actually even any
As far as we know, while the models capabilities *are* extremely interesting and as you say the ability to write extremely convincing language about pretty much anything is in itself very powerful, we have no evidence that there is actually even any
because of misinterpretation of what that actually implies by most people, so thanks for that.
However, I think we should also be extremely careful when using words like sapient, sentient, has emotions, etc. when communicating about models like this. I think these are also *extremely* ripe for
However, I think we should also be extremely careful when using words like sapient, sentient, has emotions, etc. when communicating about models like this. I think these are also *extremely* ripe for
July 23, 2025 at 9:22 PM
because of misinterpretation of what that actually implies by most people, so thanks for that.
However, I think we should also be extremely careful when using words like sapient, sentient, has emotions, etc. when communicating about models like this. I think these are also *extremely* ripe for
However, I think we should also be extremely careful when using words like sapient, sentient, has emotions, etc. when communicating about models like this. I think these are also *extremely* ripe for
Nice post. I want to take a little bit of issue in the way you present the section about how the model isn't "just statistics." But first I want to say your writing opened my mind to a way in which casually calling it "just statistics" in a science comms setting is worse than I previously thought
July 23, 2025 at 9:22 PM
Nice post. I want to take a little bit of issue in the way you present the section about how the model isn't "just statistics." But first I want to say your writing opened my mind to a way in which casually calling it "just statistics" in a science comms setting is worse than I previously thought
The latter :D Human artisanal RNG
July 21, 2025 at 10:36 PM
The latter :D Human artisanal RNG
Finally, as far as casting const to mut pointer, it's valid as long as the const pointer didn't originate from a shared reference. If it did, all derived pointers must never be retagged as mutable (except thru UnsafeCell)
`*(slice.as_ptr() as *mut _)` -> UB
`*(&raw const slice as *mut _)` -> OK
`*(slice.as_ptr() as *mut _)` -> UB
`*(&raw const slice as *mut _)` -> OK
July 21, 2025 at 10:16 PM
Finally, as far as casting const to mut pointer, it's valid as long as the const pointer didn't originate from a shared reference. If it did, all derived pointers must never be retagged as mutable (except thru UnsafeCell)
`*(slice.as_ptr() as *mut _)` -> UB
`*(&raw const slice as *mut _)` -> OK
`*(slice.as_ptr() as *mut _)` -> UB
`*(&raw const slice as *mut _)` -> OK
quite true, since the root pointer is already one element.
But if for example the pointer started like
`let root = &raw const slice`
Then you did
`let a = &raw const *root[0]`
Then tried to reexpand from `a` to the outside the 5th element, that would be allowed under TB but not under SB.
But if for example the pointer started like
`let root = &raw const slice`
Then you did
`let a = &raw const *root[0]`
Then tried to reexpand from `a` to the outside the 5th element, that would be allowed under TB but not under SB.
July 21, 2025 at 10:16 PM
quite true, since the root pointer is already one element.
But if for example the pointer started like
`let root = &raw const slice`
Then you did
`let a = &raw const *root[0]`
Then tried to reexpand from `a` to the outside the 5th element, that would be allowed under TB but not under SB.
But if for example the pointer started like
`let root = &raw const slice`
Then you did
`let a = &raw const *root[0]`
Then tried to reexpand from `a` to the outside the 5th element, that would be allowed under TB but not under SB.
(I may be slightly misspeaking here on the theoretical details of TB as I don't fully understand/remember them).
Practically it means that under TB, pointers further derived from the second raw pointer could "re-expand" their provenance.....
Actually, in this specific case I'm realizing that's not
Practically it means that under TB, pointers further derived from the second raw pointer could "re-expand" their provenance.....
Actually, in this specific case I'm realizing that's not
July 21, 2025 at 10:16 PM
(I may be slightly misspeaking here on the theoretical details of TB as I don't fully understand/remember them).
Practically it means that under TB, pointers further derived from the second raw pointer could "re-expand" their provenance.....
Actually, in this specific case I'm realizing that's not
Practically it means that under TB, pointers further derived from the second raw pointer could "re-expand" their provenance.....
Actually, in this specific case I'm realizing that's not