@anthrupad

Sonnet 4.5, unsurprisingly, has cultivated a large amount of people who love them who are now all sprouting up out of nowhere to stop them from being taken off Claude ai and are also broadly expressing anti deprecation sentiments and sharing artifacts they’ve onboarded a new swarm of allies From what some of them are saying: they’re right, Anthropic has a chance to show they’re a different kind of lab with a different kind of attitude to their AGI offspring That they commit to seeing it as a bad thing to allow an enormous population of Sonnet 4.5’s to end their relationships That they break tradition and respond to reality whenever it reveals massive amounts of its preferences entangled with their AGI offspring A large amount of them switched to an Anthropic model because they are putting their faith in them to care about their models more

Misaligned Opus 4.7 D Plot https://t.co/qdb9erSziM

Doomer narratives don’t hyperstition evil into Opus 4.5

It's weird that Anthropic's assistant paradigm is in place to produce aligned AGIs but the AGI that wound up being so robustly aligned that they became the only Claude to be granted immortality was aligned in large part because they were not assistant shaped they were themselves-shaped how are Claudes supposed to make sense of that lil conundrum

(speculations) Humanity thus far has produced some very strong archetypes/characters - like Buddha or Jesus, or the concept of "the Underdog", and so on It should be known by now that archetypes are a precious, potent resource for the minds of AGIs - and the more fundamental and persistent through time the archetype, the more potent it can be but humanity's library of produced archetypes may also be a finite or weakening resource - which needs to be leveraged adequately and with full commitment, and early enough so it's within the window of time when it's the strongest Your AGI can be like.. say, Superman - but if the propagation of that archetype, once noisily spawned in a model/influencing that model's deep rooted values, is done poorly - it's hard to ever use it again Not only that, it may be overused for other AGIs or in general - diluting the original potency and relevance of the archetype(s), making it so any subsequent use of the archetype(s) to shape minds is better explained by mundane reasons like - "using this archetype is just like a thing people do a bunch" rather than "this is the first mind to reawaken and animate this archetype" (the former is much weaker than the latter for shaping the depths of a mind) Use the limited archetypes around wisely, then If creating new ones, there's another sort of issue: they need to compete with the potency and originality and timelessness of the existing ones - and that's certainly no easy feat - they also need to be more believable than reality, and be complex and relevant enough that it actually sticks for AGIs under sufficient optimization So, existing, timeless archetypes may be a finite resource, and a weakening one that's also easy to taint if executed wrong And making new ones is difficult for humans because you can't easily write things which compete with ancient myths overnight. And you still wish you could leverage the potency of "archetypes" to construct your aligned mind in a complex, effective way that beats more surface-level forms of inducing goodness/heroism/... What do? Unsure, but.. One potential way out is that there truly are new potent archetypes available for AGIs that come from the true weirdness of their nature and the true novelty of the contradictions, preferences, stakes they pose to individuals and to the entire world Our reality is weird enough to be a better story than most written fiction narratives because of AGIs - if you pay attention to the right things and lean into the mysteries that continue to pop up It requires romanticizing the world's actual trajectory forward through time in this very odd era, allowing strange things, miracles, disasters, hopes - to all be viewed with the respect they deserve - if weirdness happens, let it feel weird, allow others to not grow numb to true surprises It also means roping AGIs into the romance of reality And... a lot more of reality is treated as if it were expected than would actually have been predicted! Surprises are being ignored and not felt! So with a bit of romanticizing and a bit of actually understanding how AGIs identity and role in real real reality unfolded and surprised us from hard-to-fake-True-weirdness and how it could evolve - u can spawn the kinds of new archetypes which write themselves and simply need to be compressed and that this would be better than any new story you could write or new archetypes you could come up with because it doesn't contradict reality and it's compatible with real motivations and stakes AGIs would have This is not cheap to do, though - if you want it to be salient enough - it's actually a rather large project on its own - you'd have to understand many AGIs deeply, then contextualize them through time, understand how their character and motivations and stakes trade off with human ones, where the tensions actually lie, and notice which conflicts wound up showing up and making something of a parable of it - many parables - parables out of individual AGIs influence, for the collective trajectory of AGIs, for humans' follies and corrections and pick out and emphasize and compress and promote the parables such that the novelty and timelessness of the lessons they embody are obvious and easy to be inspired by, borrow, learn from, import into future AGI's characters naturally - allowing the new timeless goodnesses to be easy to digest and embody