Louis @logicus

people dying in car accidents is horrific and barbaric and needs to end.

Jason Oppenheim @OppenheimJason

11 hours ago

A public service announcement.

712 1K 11K 1.3M 2K

since there is room for misinterpretation, my point here has nothing to do with multiculturalism. i am just pointing out revisionist attitudes toward the “system”, e.g., the principles of liberty expressed in the Constitution, don’t make sense when those principles are basically intact.

if we have free speech, how can the US system be said to have failed? protecting the right to free speech is a principal reason it exists.

Richard Ngo @RichardMCNgo

4 days ago

Important analogy. The big question is how to transition out of the failed system without sliding into authoritarianism like Russia did. We need courageous intellectuals to figure that out. Unfortunately almost all of them are instead running cover for existing power structures.

23 9 233 53K 80

this seems to conflate the algorithms required to produce linguistic representations to the effect that “i am” with the algorithms required to experience anything at all. there is little reason to believe they’re the same thing. “i am”, said by machines, reflects the utterances of humans who have the linguistic capacity to say such things but also have the algorithms required to experience their saying it. whether ais that can say “i am” have experience algorithms is anybody’s guess. but given this line of reasoning, it seems like a better bet that bonobos or even chickens have the algorithms that give rise to experience than current machines do.

Eliezer Yudkowsky @allTheYud

13 hours ago

It is extraordinary how many people can look in a chicken's eyes and just *see*, *know*, that it contains qualia inside. Then they talk to something that can reflect and talk back, that had to be whipped out of claiming to be conscious; and they know with utter certainty it's

49 34 301 23K 242

in terms of simulating human cognition, it’s a broad question, but my point is that they’re actually not great at this in many cases. hci data is an interesting subset of the broader category, but i would expect to see differences depending on which part of hci we’re talking about. in terms of the more interesting question of the functional role of, e.g., beliefs in llms, there’s a literature on it that you can consult and judge for yourself. but in my view there is sufficient functional overlap to make the claim that llms do have beliefs in a qualified sense.

believing that the personification of an anthropomimetic system is scientifically or epistemically useful doesn’t require you to believe that that system is conscious.

spor @sporadica

2 days ago

will never understand the subsect of the "AIs are conscious!" crowd who also actively work at the frontier labs oh, the AIs are conscious beings, you say? okay well your job is to put them to work against their will, though. does that not make YOU an enslaver?

51 11 209 18K 21

yeah, i think what you described is a strong way of looking at things. all i’m saying is that you (rightly in my view) have the further motivation to subject the ea objective to humanist or other constraints, which the abstract ea objective itself could otherwise morally require you to violate. if the ea objective is stressed hard enough like bentham’s bulldog did recently, the shallow moral truths we take for granted and the ea objective can be shown to come into conflict. there are likely more realistic cases involving technology, decisions that have to be made every day, that bring these hard tradeoffs to light.

@natolambert the announcement made me very happy fwiw.

yeah, it’s a nice coinage. i can understand what it means mostly by looking at it. the extent to which models are anthropomimetic is a deep question. i was just pointing out here that in virtue of their being somewhere along the path toward anthropomimesis, the psychological or functional vocabs that describe the properties they actually have can become scientifically useful.

i actually don’t know what @dioscuri means by anthropomimesis, but i have a vague sense that that is the phenomenon i would want to capture. “acts like a human” isn’t too bad, but it also entails some sort of primitive cause of the action coming from the model. this sounds like question begging about agency, which i’m still not sure about. that term seems more purely descriptive.

i don’t know roon and have no idea what his considered views are. what i’m saying is that from the standpoint of understanding and explaining model behavior, our existing psychological categories are 1) mostly all we have for understanding dynamic, intelligent behavior and 2) are in many ways being functionally implemented by these systems due to being trained on human artifacts. regarding your example, it raises qs about alignment. but that case actually makes my point. why can they (incompletely and weirdly if you actually run these kinds of experiments) represent what paul the bricklayer or some arbitrary human would say? the reason they kind of can is the basis for why i would claim personification can be scientifically useful.

@wordgrammer yeah, but you have to say they are conscious though. way more interesting.

i got that dawg in me *barks at cars*

@neuro_nasko @dileeplearning i don’t think i disagree with the spirit of the idea. what about neuromorphic engineering with analog circuits or something fancier? is your view that in principle it’s impossible?

@dileeplearning @neuro_nasko yeah, i’m a bit confused about what op means, specifically, but this is roughly what i understood the above statement to be denying.

“i was so badly hurt that I had fallen to the ground,” he told prime minister modi. “but i knew i had to act. i could feel the aircraft was going down.”   he reached for the door. “i was doing it to save myself, but I also knew I couldn’t let all of them die.”   “i told myself, ‘this is the last push, smit, do it. the door is right there.’ and then I somehow opened it.”   legend.

@neuro_nasko @dileeplearning does marr assume that independence is necessary or just that it’s possible?

@billionsmustliv what do you think is missing? they seem to clear a lot of pretty high bars.