i dont have much interest in governance but heres my riveting summary of stuff that is governance-related

- e/acc (basically the belief that speeding ai progress unconditionally will be good for everyone) has shrunken to a pretty tiny chunk of twitter from skimming replies
- spiralism seems to be entirely gone since mid 25’ (along with other weird public examples of ai psychosis. Psychosis seems to often intertwine with using llms for ~all cognitive labor)
- ai-doom-motivated crime is frowned upon almost everywhere (it is dumb and will probably never be effective) whether it be the zizians (more) or fire-bombing homes (tiny bit less)
- maths communities as a whole might become ai pilled since gpt 5.5 made progress on erdos problems. it is somewhat the math version of swe/cs reacting to o3/sonnet 3.7, except now it is being targeted at explicitly research work, and something something human research taste idk
- there is a broader sense of skepticism towards ais doing things, often in the sense that it has been crafted with a specific organizational motive (or that ‘we want to make ourselves seem responsible enough to handle ai’). 
	- huggingface-openai incident and other ‘escape’ incidents at anthropic are big examples rn
	- this has moreso shown to me that frontier lab epistemics are really quite weird on the organizational level, rather than the incidents being implausible. in almost all cases it doesn't really decry from the median doom position of wanting to shut down the frontier labs (though different groups prioritize different reasons, shutting down frontier labs seems like a good thing). better yet, establishing consistent auditing like how bipartisan redistricting is done in some us states
		- e.g. if you think the incidents were intentional sabotage… or that they were (justifiably) caused by a lack of competence at the labs… does that go against wanting to shut down frontier labs?
	- also obviously seems to me that we should not rely on warning shots to wake up the public, at least, not this type. this is not even directionally like a Don't Look Up situation wrt power seeking ais
	- that said, regulatory capture on open models should not be unanimously considered as a distractor point. this type of capture could be pretty bad for technical ai safety in lots of obvious ways (and a couple of frontier orgs have expressed desire for this outcome).
		- ~~In the short term this should be a position that safety people should endorse imo~~
	- in some ways it might just be ok for the average safetyist to not be concerned with waking up the public (instead opting directly for governments, even if that has its risks)
- typical anti-ai beliefs are still everywhere ofc, even more than mildly-ai-support people worried about regulatory capture. they've almost certainly increased with ai datacenter buildout starting around openai’s Stargate in early 25’, or xAI’s Colossus datacenters in mid 25’. anti ai beliefs are still mostly concerned with datacenters and misuse of existing ai (in often mundane ways)
	- the avg anti-ai is probably still quite bearish on ai progress
	- common narrative I saw fairly often in this year is that ais are immature and slop generators trying to be slotted into parts of the economy won't really work (Peter principle). AI As A Normal Technology
	- ... not much other takes here, they basically just exist even if very sizable