Nathan Hoobler
@nathanhoobler.bsky.social
Pixel Pusher and Digital Illusionist
@ Nvidia
All opinions my own
@ Nvidia
All opinions my own
Shane I have been forced to learn way too much about catholic integralism over the past ~10 months and I am not happy about any of it.
December 13, 2025 at 4:05 AM
Shane I have been forced to learn way too much about catholic integralism over the past ~10 months and I am not happy about any of it.
An inevitably two-party first-past-the-post system like the US is unfortunately saddled with literally cannot function without them!
December 13, 2025 at 2:46 AM
An inevitably two-party first-past-the-post system like the US is unfortunately saddled with literally cannot function without them!
"Anarcho-libertarian, but from a totalitarian communist perspective"
December 12, 2025 at 12:40 AM
"Anarcho-libertarian, but from a totalitarian communist perspective"
Not an expert, but I imagine CM without corresponding evasion probably isn't very effective, and they were already in a low-energy state.
December 5, 2025 at 7:06 PM
Not an expert, but I imagine CM without corresponding evasion probably isn't very effective, and they were already in a low-energy state.
the threatening aura of that switch labeled "Use Regex"
December 5, 2025 at 3:03 AM
the threatening aura of that switch labeled "Use Regex"
I suppose the text mediates the expression of some sub-textural experience within the author into a separate sub-textual experience within the reader?
December 4, 2025 at 7:52 PM
I suppose the text mediates the expression of some sub-textural experience within the author into a separate sub-textual experience within the reader?
The received interpretation of the reader and the actual indented meaning of the sentence as stated by the author are distinct but ideally related.
December 4, 2025 at 7:52 PM
The received interpretation of the reader and the actual indented meaning of the sentence as stated by the author are distinct but ideally related.
Personally I'm *very* comfortable with MDs using AI to diagnose, just not with them *only* using AI. A system which could flag potential signals from imaging, for example, for secondary confirmation or rejection seems like all upside.
December 4, 2025 at 7:47 PM
Personally I'm *very* comfortable with MDs using AI to diagnose, just not with them *only* using AI. A system which could flag potential signals from imaging, for example, for secondary confirmation or rejection seems like all upside.
(This is not to say I think brains use the same vector representation or activation function approach necessarily, but the overall effect seems to "rhyme")
December 4, 2025 at 7:38 PM
(This is not to say I think brains use the same vector representation or activation function approach necessarily, but the overall effect seems to "rhyme")
If I weren't unnaturally aware of my cognition here (because of the context of this conversation) I would have just given my first interpretation and been done with it.
December 4, 2025 at 7:38 PM
If I weren't unnaturally aware of my cognition here (because of the context of this conversation) I would have just given my first interpretation and been done with it.
... But if a few possibilities in the space of interpretations are even slightly more likely, they're maximized and the remaining terms will trend towards zero. Even though I've cited ambiguities all along my brain pretty quickly started assuming the most likely interpretation as the right one.
December 4, 2025 at 7:38 PM
... But if a few possibilities in the space of interpretations are even slightly more likely, they're maximized and the remaining terms will trend towards zero. Even though I've cited ambiguities all along my brain pretty quickly started assuming the most likely interpretation as the right one.
So I think in this sense there's actually a pretty good analogy from AI with normalization and activation functions. In a case where outputs at a certain layer all have equal weight, they are treated as equally likely and ambiguity dominates.
December 4, 2025 at 7:38 PM
So I think in this sense there's actually a pretty good analogy from AI with normalization and activation functions. In a case where outputs at a certain layer all have equal weight, they are treated as equally likely and ambiguity dominates.
I'd say it makes me more certain in the most likely of my original interpretations by being consistent with other circumstances where I'd seen this type of statement.
December 4, 2025 at 6:52 PM
I'd say it makes me more certain in the most likely of my original interpretations by being consistent with other circumstances where I'd seen this type of statement.
I'd say my thinking is informed by the prior context including subject matter (specific things like "committing fraud" but also kind of general subtle tone elements? The context is sort of difficult to parse grammatically so I am falling back on "vibes", really.
December 4, 2025 at 6:47 PM
I'd say my thinking is informed by the prior context including subject matter (specific things like "committing fraud" but also kind of general subtle tone elements? The context is sort of difficult to parse grammatically so I am falling back on "vibes", really.
I am thinking about the entire history of my previous experience with language and this particular phrasing. How often each interpretation occurs, and which is likely to occur in general circumstances versus very special cases. I'm trying to correlate this with a cluster from my prior experience.
December 4, 2025 at 5:51 PM
I am thinking about the entire history of my previous experience with language and this particular phrasing. How often each interpretation occurs, and which is likely to occur in general circumstances versus very special cases. I'm trying to correlate this with a cluster from my prior experience.
So based on my experience, there's vanishingly small chance "Eddie" is not the subject.
By default I'd assume 80% of the time it's referring to his emotional state. Previous sentences in the conversation would possibly contain information about whether his emotions are in a state that matches that.
By default I'd assume 80% of the time it's referring to his emotional state. Previous sentences in the conversation would possibly contain information about whether his emotions are in a state that matches that.
December 4, 2025 at 5:43 PM
So based on my experience, there's vanishingly small chance "Eddie" is not the subject.
By default I'd assume 80% of the time it's referring to his emotional state. Previous sentences in the conversation would possibly contain information about whether his emotions are in a state that matches that.
By default I'd assume 80% of the time it's referring to his emotional state. Previous sentences in the conversation would possibly contain information about whether his emotions are in a state that matches that.
OK!
Without any context I'd say it is *very likely* an imperative directed at an individual named "Eddie" to either reduce his expression of some strong emotion he's currently experiencing or to commit to some specific life choice?
Without any context I'd say it is *very likely* an imperative directed at an individual named "Eddie" to either reduce his expression of some strong emotion he's currently experiencing or to commit to some specific life choice?
December 4, 2025 at 5:23 PM
OK!
Without any context I'd say it is *very likely* an imperative directed at an individual named "Eddie" to either reduce his expression of some strong emotion he's currently experiencing or to commit to some specific life choice?
Without any context I'd say it is *very likely* an imperative directed at an individual named "Eddie" to either reduce his expression of some strong emotion he's currently experiencing or to commit to some specific life choice?
That's fair. I guess I'd say they are good at producing mediocre translations, and bad at producing good translations?
December 4, 2025 at 5:06 PM
That's fair. I guess I'd say they are good at producing mediocre translations, and bad at producing good translations?
Also, and I mean this in he most open "I don't know the answer" way:
Does a blind person understand the concept of color?
Obviously they could talk about it, but it would ONLY be known to them via linguistic communication from others, right?
Does a blind person understand the concept of color?
Obviously they could talk about it, but it would ONLY be known to them via linguistic communication from others, right?
December 4, 2025 at 4:47 PM
Also, and I mean this in he most open "I don't know the answer" way:
Does a blind person understand the concept of color?
Obviously they could talk about it, but it would ONLY be known to them via linguistic communication from others, right?
Does a blind person understand the concept of color?
Obviously they could talk about it, but it would ONLY be known to them via linguistic communication from others, right?
December 4, 2025 at 4:46 PM
Also, and I mean this in he most open "I don't know the answer" way:
Does a blind person understand the concept of color?
Obviously they could talk about it, but it would ONLY be known to them via linguistic communication from others, right?
Does a blind person understand the concept of color?
Obviously they could talk about it, but it would ONLY be known to them via linguistic communication from others, right?
Could you expand on that a little bit? What do you mean by "between the words and reader/writer"?
December 4, 2025 at 4:36 PM
Could you expand on that a little bit? What do you mean by "between the words and reader/writer"?
I'm pretty hesitant to talk about LLMs "knowing" anything or "thinking" in ways we would identify in part because they have a fundamentally different relationship with time. At best you might imagine them as perpetually unconscious, flashing reactions when poked by a query.
December 4, 2025 at 4:35 PM
I'm pretty hesitant to talk about LLMs "knowing" anything or "thinking" in ways we would identify in part because they have a fundamentally different relationship with time. At best you might imagine them as perpetually unconscious, flashing reactions when poked by a query.
I'd agree with that (with the amusing caveat that "language translation" ironically seems like something they are consistently good for -- it's all the rest of the human stuff they fail at).
I think the real question is whether they would converge to better output with more/better training or not.
I think the real question is whether they would converge to better output with more/better training or not.
December 4, 2025 at 4:32 PM
I'd agree with that (with the amusing caveat that "language translation" ironically seems like something they are consistently good for -- it's all the rest of the human stuff they fail at).
I think the real question is whether they would converge to better output with more/better training or not.
I think the real question is whether they would converge to better output with more/better training or not.