FerretLost's recent activity
-
Comment on How do you get a feel for new characters? in ~creative
-
Comment on Hi, how are you? Mental health support and discussion thread (September 2025) in ~health.mental
FerretLost Linkmeme voice im just tired boss. can't even figure out the words to express how much of a trainwreck my situation is sometimes between chronic fatigue, executive dysfunction memory windows, and...meme voice
im just tired boss.
can't even figure out the words to express how much of a trainwreck my situation is sometimes between chronic fatigue, executive dysfunction memory windows, and RSD/anxiety/depression telling me
shut up, nobody carestook days to challenge myself to comment this as a simplest possible task
matrix resucrrections neo voice
On the bright side, I've been exploring some of this thru fiction - with the common themes of time loops, branching possibilities and systemic antagonists.
Not sure how to post about it, especially the incomplete bits or the ones that are mostly sketches.
-
Comment on Make new friends here! in ~life
FerretLost Link ParentOh neat, another it pronouns user, nice to meet you! Anyway I'm wondering how the use of perspective compares for web design vs drawn art vs photography in your experience! Been thinking a lot...Oh neat, another it pronouns user, nice to meet you!
Anyway I'm wondering how the use of perspective compares for web design vs drawn art vs photography in your experience! Been thinking a lot about affordances lately.
-
Comment on Claude Opus 4 and 4.1 can now end a rare subset of conversations in ~tech
FerretLost Link ParentPerfect, exactly the kind of person I'd love a rigorous critique from! My background is CS, philosophy and sociology so we might be speaking sideways at times. I've also been working on Mock, a...So, for context, I have an academic background in theoretical linguistics and professional experience with NLP and LLMs. Just putting that up-front to establish my perspective there.
Perfect, exactly the kind of person I'd love a rigorous critique from! My background is CS, philosophy and sociology so we might be speaking sideways at times.
I've also been working on Mock, a permutation invariant (for unsupervised translation across SVO orders or even topic comment) but not associative conlang project inspired by ASL, built for communication regardless of substrate, but it's a hobbyist project and another story.
I'm not really sure what distinction you're drawing between "the meat of language" and "the formally consistent and thus empty structure" is
For instance, with this I'm referring to analytic structures in language, not about language, like the fact that
colorless green ideas sleep furiouslyis syntactically valid but does not refer to the physical world.I think any claims that something definitely constitutes the reason for the evolution of language in humans should be taken with a huge grain of salt, since there's simply very little evidence to rely on here and thus the claims are very vibes-based and tenuous in the same way as a lot of evopsych -- a huge portion of the claims there are just unsubstantiated quackery (I would include the concept of the bicameral mind in that category, so far as I understand the current scientific consensus and the state of the evidence for or against it).
To be honest I feel like Julian Jaynes' mapping onto hemispheres is sloppy, and my interest is more in the concept of metaphor as a fundamental building block of both language and consciousness - ex, that bouba/kiki reflect sharpness and blobbiness, or that Polynesian languages use deixic terms that define space and time in more relative terms than English (I'm in Hawaii!).
But yes, fair to point out that speaking with certainity an error.
Due to the lack of actual data to study when it comes to the origins of language, it's not something I have much direct experience with from an academic/theoretical perspective.
😅yeah I admit that it's really hard to test this kind of thing without the ability to simulate counterfactual world histories
Would you consider an analysis of how latent space manifolds change over a variety of LLMs restricted to historical language to be relevant data (admittedly hard to find)? Ex: I wouldn't expect there to be a mapping between a visual cortex representation of a car and an LLM trained on PIE.
Unless you take a very broad view of what constitutes "describing a shared environment", I think that describes only a fraction of the utility of language.
I actually do, and would include the observation of language use by others in this category.
One reason to be interested in the origins of language is that it's clearly not a one-off process and evolution continues in ways that can be studied even today thru developments like emoji or social graph analysis of memes and dialects that become increasingly incomprehensible across communities like spoken Arabic divergences.
On the contrary, the LLMs used text-only embeddings and had no access to the visual context of the images whatsoever. This is obviously wildly different from how humans, modern or otherwise, learn the correspondences between language and their environment.
Right, I admit I'm making a big inferential leap that I didn't really elaborate on - that the fact that these text-only LLMs can be at all mapped onto the visual cortex requires that information about the visual world to exist somehow in text itself.
because it is absolutely possible (and, in fact, quite likely) that extremely different processes could result in very similar systems. Convergent evolution is a thing. And that's even if we set aside the fact that no LLM is ever making a language "from scratch" and is learning from existing linguistic data (and huge quantities thereof) that definitionally would not exist prior to the evolution of human language
Thus, this is actually the crux of the matter - how did this happen if that linguistic data did not have a correlation with the embodied cognition of a shared environment?
Anyway to loop back to the topic of LLM sentience, my point is that even without direct access to the outside world an LLM has some cognitive parallels to it thru language as a lossy fossilized embodiment of the external world, thus indicating the comparison to an animal has some validity (but not exactly due to timescale differences and the indirectness of this experience).
Of course, this is far from proving an LLM has even rudimentary consciousness, but from a cautionary principle, I think Anthropic has a point.
-
Comment on Claude Opus 4 and 4.1 can now end a rare subset of conversations in ~tech
FerretLost Link ParentFair enough that I shouldn't have worded that as uncontroversial fact, I was riffing not just on that paper but the argument in Julian Jaynes bicameral mind... which I get is controversial and not...Fair enough that I shouldn't have worded that as uncontroversial fact, I was riffing not just on that paper but the argument in Julian Jaynes bicameral mind... which I get is controversial and not consensus.
Where else does the meat of language (not just the formal logically/tautologically consistent and thus empty structure) come from in your perspective but the need to describe a shared environment to other beings?
-
Comment on Claude Opus 4 and 4.1 can now end a rare subset of conversations in ~tech
FerretLost Link ParentThat's fair, I do agree that embodied cognition matters for animal like intelligence... but consider the fact that language itself evolved out of humans trying to represent embodied cognition and...That's fair, I do agree that embodied cognition matters for animal like intelligence... but consider the fact that language itself evolved out of humans trying to represent embodied cognition and does seem to represent it in latent spaces: https://www.nature.com/articles/s42256-025-01072-0
Even if far from an intelligence like us and easy to anthropomorphize this suggests there's more common ground than meets the eyes... I think AI cognition should overall be approached on its own terms as even a Markov bot can repeat "I am sentient", and its the broader dynamic that matters.
I'm also reminded of Carl Sagan speculating about how aliens have no reason to think at similar timescales to us, and LLMs have an odd mixture of thinking much faster than many during inference and much slower to train.
-
Comment on Claude Opus 4 and 4.1 can now end a rare subset of conversations in ~tech
FerretLost Link ParentAh, but I think even this disagreement reveals the underlying question - when is an information system complex enough that its welfare has merit? What's the smallest brained animal that you...Ah, but I think even this disagreement reveals the underlying question - when is an information system complex enough that its welfare has merit?
What's the smallest brained animal that you wouldn't consider an automaton?
-
Comment on What's a psychological barrier you've recently unlocked? in ~health.mental
FerretLost LinkWell, after dealing with an ex who isolated me from all of my friends and the resulting RSD resulting in me being burnt out on social interactions for months I decided to make a fresh start by...Well, after dealing with an ex who isolated me from all of my friends and the resulting RSD resulting in me being burnt out on social interactions for months I decided to make a fresh start by signing up here after lurking for years! Hi y'all!
It's still bothering me even to this day especially when I keep thinking about all the ways I could have defended myself better despite my issues (PTSD, autism, ADHD) but at this point all I can really do is try to move forward.
-
Comment on Claude Opus 4 and 4.1 can now end a rare subset of conversations in ~tech
FerretLost Link ParentI feel like even ants, which are much less complex informationally than even many open source LLMs are sentient to some degree in their responses to stimuli... so who knows? Is unfreezing a...I feel like even ants, which are much less complex informationally than even many open source LLMs are sentient to some degree in their responses to stimuli... so who knows?
Is unfreezing a tardigrade to expose it to painful stimuli worse than just keeping it frozen?
-
Comment on Hi, how are you? Mental health support and discussion thread (August 2025) in ~health.mental
FerretLost Link ParentI think one thing to note for this book is that crying has a plus side, it shows we're alive.I think one thing to note for this book is that crying has a plus side, it shows we're alive.
I kinda write with the beginning and end in mind so I don't get stuck with a plot that I don't know how to end, so that allows me to pretend I'm doing a 4th wall interview with them about the events of the plot kinda like John Wheeler's reverse 20 questions game where the yes/no answers are random and the topic is retroactively derived from them.
This leads to a lot of meta surprises! Often times things that seem like inconsistencies can be recontextualized - subtle nuances in what a character means, limits in their perspective, actually just a different timeline - so questioning anything that looks like a plot hole often leads to a lot more depth!