tbabb
banner
tbabb.bsky.social
tbabb
@tbabb.bsky.social
advanced cat befriender
@tr_babb on twitter
Oakland, CA
truth
November 27, 2025 at 3:07 PM
bagck in blagck
November 27, 2025 at 4:32 AM
do people think I'm advocating for intellectual isolationism?
November 26, 2025 at 10:41 PM
I have to admit I don't know why this seems to make people mad
November 26, 2025 at 10:31 PM
Yeah but I still want it to fail

I make no claim about how prevalent earnest interaction is (though I would guess it's more common outside of tpot in slopland), merely that earnest interaction with it is inadvisable
November 26, 2025 at 8:47 PM
lol, nice

fwiw I don't object to "fucking with grok" style interaction; I would lump that in the same category as jokes
November 26, 2025 at 8:22 PM
oo, do you have a snippet of it doing that?
November 26, 2025 at 8:09 PM
fair enough
November 26, 2025 at 8:03 PM
and also, I assume (at least for now) that grok's context window does not extend outside its current thread
November 26, 2025 at 8:02 PM
I kind of assume that with the data scales needed to train an LLM they would be leveraging all the data they conveniently have access to and using Everything
November 26, 2025 at 8:02 PM
is that true? i.e., does talking to grok directly influence its training with higher weight than just discoursing normally on the website?
November 26, 2025 at 8:01 PM
hm, concretely, how does using grok as an information source increase your power?

certainly I buy that playing the discourse game on twitter in general has advantages (obviously I think so bc I'm still there!), but grok seems like a bid for a new minigame that loses if everyone ignores it
November 26, 2025 at 7:58 PM
but actually I might trust fox news more than grok, since grok isn't a primary source for anything
November 26, 2025 at 7:44 PM
grok is in the same category for me. and I feel kind of the same about people who ask grok about stuff as people who would consult those questionable sources to learn what's going on in the world
November 26, 2025 at 7:44 PM
like, I feel I have a pretty good model of how truth social or alex jones or fox news are motivated and prone to lie; also they tell the truth sometimes; also they are quite impoverished sources of insight. I would basically not consult them except to see what info others are getting from them
November 26, 2025 at 7:44 PM
yeah, I don't think that grok is epistemically hazardous really (it's pretty plain what's going on) just a bad faith game player, and also stupid on top

the stupidity actually makes it less dangerous. but IMV just really degraded source of information no matter how you slice it
November 26, 2025 at 7:39 PM
maybe it's the case that "win the game you've been placed in" is the best move available to you, in which case fine (maybe) as long as you also account for the costs of things like "signaling to others that you can be manipulated by games"
November 26, 2025 at 7:12 PM
yeah, the winning countermove to "being captured in someone else's game" is usually not "win their game" but "create a new game"
November 26, 2025 at 7:10 PM
true. if other models performed similarly, then my response would be more towards "adjust my expectations" than "this thing eats glue"
November 26, 2025 at 6:54 PM
I'm wary about this reasoning. powerful person creates a game (i.e. competing for influence), people with less power try to win the game that's been created. trump does this and it seems obvious to me that playing the game at all gives him much more power than succeeding at gaming his attention
November 26, 2025 at 6:53 PM
I must be missing something because I can't comprehend how grok is considered anywhere near in the same league as the other frontier models
November 26, 2025 at 6:48 PM
I think part of my willingness to write grok off completely is that not only is it compromised, it is also useless. so there's kind of no excuse for listening to it
November 26, 2025 at 6:47 PM
maybe it gets better when paid (?), but my experience with grok has been to ask it to explain a tweet every so often to gauge its perception, and every single time it falls completely on its face. other models usually nail the same tweets, and without the context grok has access to
November 26, 2025 at 6:47 PM
I wish this would hurry up and resolve, I'm trying to outsource all my gpu workload to betting markets, and I have a lot more of these to get through
November 25, 2025 at 1:16 AM