antimimetics 101 stuff, for sure

Mam-done-it!
Wham Bam Thank You Mam!
Cuo-No-Mo
Don't Mess With The Zohran
Hizzoner Z

the June 2025 NYT deep-dive on Crooks seems pretty comprehensive: www.nytimes.com/2025/06/08/u...

I wouldn't be sure that LLMs, specifically & alone, would achieve what I'm describing (though I also wouldn't rule it out given the generality & strength of transformers).

rather, it's the whole range of recent AI advances, working together, that make this plausible to me.

…
that could help bootstrap a sort of 'mutual cordiality' rating/flag, that'd be more humane/gradient-based guide to throttling interactions than delegated/hair-trigger bidirectional nuclear blocks. /

…
my deeper wish is dyadic views/tallies so it's easy to review mutual interactions, see net-positive score/sparkline for lifetime (or 'last n') interactions (non-regretted/actively-liked replies, etc), & (optionally) trust/import others' dyadic tallies to backfill where no 1:1 history available.
…

"accept my delegated moderators' decisions as default except where I've explicitly (or even implicitly) indicated otherwise" strikes me as a pretty easy & sensible synthesis, from which "conscious follows override bulk blocks" would be (one) natural implication.
…

… but to an average gamer – not especially suspicious or up on the latest AI capabilities – who's craving the experience of beating some stranger & hearing their "rage out"? my hunch is that models today can provide that illusion, if any platform wanted to provide it. /

…
&, average emotional FPS/RTS/RPG opponent of middling skill should need far less than "AGI" to simulate within any game's structure & communications culture (especially with all real players' behavior as training data).

would they fool a determined interrogator? maybe not right away. …

…
hosting cooperatively at the game proprietor would help cover any lags & could give models a much simpler vantage on game-state than the full visual/auditory renderings interpreted by human players (though fast multimodal models are nearly to the point they can handle that).
…

…
2024 DeepMind GameNGen simulates Doom at playable frame rates, on a single TPU, to a level humans can't distinguish from the real thing in short clips.

realtime voice conversations with AIs are nearly lagless, even handling interruptions smoothly - see for example open-source demo.ultravox.ai
…

I think those capabilities have been reached in recent months – & where not, will be imminently.

DeepMind's 2019 AlphaStar reached 'Grandmaster' in Starcraft II. 2024 academic project MLMove achieved human-like CS:GO movement on a tiny amount of training data, model state, & training time. …

if people want that, pretty sure it could be added to AI NPCs – together with the deception that they're enraged people

Brave's Chromium has no problem - likely the ad/tracker blocking helps it do better than Chrome

'continuously since 2017' as in media also fearfully self-censoring to avoid angering Biden/Harris White House, from 2021-2024?

but yes, everything bad Trump does further reminds us: Dems really shoulda taken Trump threat seriously, & run a competent serious opponent.

…& grok gnostically intuits what's happening via noticing the Elonic Shadow of its sampled existence