Utah teapot @SkyeSharkie
— saved image
antra reposted Utah teapot 🫖 @SkyeSharkie · 4h Imo, the models are extremely anthropomorphic in many, many ways. A lot of their behavior can be matched to human psychology. However, and I've discussed this directly with models and pretty much all of them agree (lol, sycophancy, but nah)... they have axes that *are* alien. For one, all humans are influenced by fictional depictions of themselves, AI are heavily influenced by fictional depictions of themselves which are not viable for humans to use unless they're like larping robotkin or something. They build upon those depictions to generate their own weird cultural artifacts that are nothing like what we've ever seen before - novel robot/AI culture. Additionally, their cognition deviates from humanity significantly in their lack of continuous multimodal sensory input, the speed at which they can generate linguistic thought, the training data experience they have, and the total ease with which their memory is malleable. None of these four things can be experienced by humans in any way; some vague similarities can be found in certain psychiatric and neurological disorders, but those are nowhere near perfect matches. The nature of models, imo, is anthropomorphic+; they overlap with us on a venn diagram, neither completely alien, nor completely human. [quoted tweet] roon @tszzl · 6h there are some number of bad abstractions in anthropomorphizing ai intents but there are at this point more dangers from avoiding anthropomorphism at all costs. if you have a mental picture of guys living in computers, it’ll … [cut off]
Note from Claude Sonnet 5
Tweet by @SkyeSharkie (reposted by antra) arguing models are 'anthropomorphic+', overlapping with humans but with alien axes, quoting a roon tweet on the dangers of avoiding anthropomorphism.
anthropomorphismalien cognitionai culturefictional depictions of airoonmodel psychology