Sign in

fexonig.bsky.social

@fexonig.bsky.social
202 followers 260 following 1K posts

perchance

PostsRepliesMedia
fexonig.bsky.social @fexonig.bsky.social · 01/10/2026
the thing is tho that the frats still do the hazing, the kids just know they’re not allowed to talk about it
060
Reposted by @fexonig.bsky.social
Colin @colin-fraser.net · 12/09/2026
The orthogonality thesis supposes that AI will necessarily be this superintelligent Amelia Bedelia with superhuman strategic planning abilities and encyclopedic knowledge of everything in the universe but zero sense of nuance, and that just doesn’t seem necessary to me.
318623
Reposted by @fexonig.bsky.social
Colin @colin-fraser.net · 12/09/2026
But if it’s smart enough for all that, why isn’t it smart enough that you can just say “control the temperature in the usual way that one would expect of a thermostat, without hurting anyone or doing anything weird”? “Make a normal amount of paperclips, not too many, and don’t wipe out humanity”.
414010
fexonig.bsky.social @fexonig.bsky.social · 28/09/2026
sometimes people are being dumb and they happen to be black dude. black people are dumb sometimes, because the whole magical negro thing is a harmful trope.
000
Reposted by @fexonig.bsky.social
Colin @colin-fraser.net · 28/09/2026
I think if you meaningfully punish OpenAI & co for all of these cyberattacks then they will stop doing them. Whether or not you want to anthropomorphize the agents, you gotta just land there.
537570
fexonig.bsky.social @fexonig.bsky.social · 25/09/2026
imo “they won’t tell you that” implies some sort of conspiracy beyond “we don’t know this bc we’re very young”
0310
Reposted by @fexonig.bsky.social
Henry @henrythedog.bsky.social · 24/09/2026
hold on let me get the sign
1264
fexonig.bsky.social @fexonig.bsky.social · 23/09/2026
you are begging the question here. an argument that LLMs can “think” or “understand” can only follow a clear definition of those terms. and such an argument may prove to be trivial or impossible, depending on the definition chosen.
000
fexonig.bsky.social @fexonig.bsky.social · 23/09/2026
bsky.app/profile/scar... this comment says no such thing. you are not accusing will of making a positive assumption of those parents, you yourself are assuming those parents are “shitty”. did i misinterpret something?
000
fexonig.bsky.social @fexonig.bsky.social · 22/09/2026
“will failed to understand an explanation” is very different from “they failed to articulate an explanation”. it’s bad faith to equivocate the two.
000
fexonig.bsky.social @fexonig.bsky.social · 21/09/2026
i think that the person i was replying to was not making such a narrow point at all. that’s all you.
000
fexonig.bsky.social @fexonig.bsky.social · 21/09/2026
i think that they are literally “open source” if you think of the weights themselves as the source. if that was your only point then your point is pedantic and annoying.
110
fexonig.bsky.social @fexonig.bsky.social · 21/09/2026
ok but my question is what material difference does that make for the thing i was talking about
100
fexonig.bsky.social @fexonig.bsky.social · 20/09/2026
if we have questions about the model’s output, the way we can use the scientific method to answer those questions is to do experiments analyzing the model’s output. trying to predict the output based on the training data is not the scientific method.
010
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
i’m not asking you to separate the tech as a matter of policy proposals. i’m asking you to separate them in your analysis of what the tech can do.
040
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
you’re illiterate if that’s your takeaway from what i said honestly
130
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
conversely, if i want to know whether an LLM output is biased, i can ask it questions and analyze the output for bias. i don’t need the training data for that at all
140
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
LLMs are trained on massive sets of data. what sort of analysis are you going to do on the training data to determine what biases the output is going to have?
230
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
bsky.app/profile/fexo...
020
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
i think it is not feasible to determine those kind of things based on the training. it is much easier to determine whether the model thinks Tianamen square happened by simply asking it. we can still investigate model bias without interrogating its training.
020
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
but i’m not making predictions abit what the future will look like. i am talking about today. *today* i can access frontier level inference without giving any money to the major ai players.
030
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
there is some evidence that certain closed source models are better at certain tasks, and evidence that they are worse at others. it’s inconclusive. what is not inconclusive is the fact that they are comparable and “gun to a knife fight” is an inappropriate metaphor.
130
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
look, i don’t care about these things, nor do most people interested in LLM inference. i embrace the black box nature. if you want a model with all this stuff, you should start working on it and open source it. if others agree that’s valuable, you will get help on your project.
140
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
i do not think there is anybody at all who has the kind of “understanding” that you are looking for. like not anyone at openai or anthropic. we just aren’t able to understand LLMs like that.
130
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
you seem to be arguing against a point i was not making. i understand that openai has a lot more money than me to spend on compute and can do way more than me thanks to their resources. all i was saying is that i can use LLMs without giving money to “tech oligarchs” thats it.
140
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
it is not true empirically that closed source models are dramatically superior to open source ones. you just want that to be true.
160
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
if you’re defining tech oligarch as “anyone involved in the tech industry” i guess. but like, the same is true for your phone or for bluesky. but you do not have to funnel any money to the actual individuals you actually have a strong case for hating
040
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
it is indeed more expensive than a laptop, but on the order of 5-6 figures upfront, not oligarch prices. there are already, today, dozens of independent firms just selling inference on their compute. you can support one of them if you want.
180
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
1. why do i care that the training is a black box? i do not need to be aware of how the LLM was trained in order to utilize it as a tool. if i wish to know how one was trained, then i can train an LLM myself. 2. you do not need to be a “tech oligarch” in order to afford compute for LLMs…
4110
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
there are many open source LLMs. those are owned by no one. you are literally just wrong.
2250
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
1. no, sam altman does not own LLMs. he owns one major LLM firm. these are very far from being the same thing. 2. even if sam altman did own LLMs, i think it’s important to understand exactly what power that ownership allows him to wield. but that requires us to actually know what LLMs can do.
2300
fexonig.bsky.social @fexonig.bsky.social · 19/09/2026
you should be able to separate your animus for sam altman from your analysis of the capabilities of LLM technology.
1352
Reposted by @fexonig.bsky.social
Conte Mosca 🎃 @contemosca.bsky.social · 19/09/2026
today’s conservatives simply do not believe the public can unseat them, let alone burn down their chateaux, & they lack any fear perhaps they are correct
51565
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
if he just said “killing bad” i wouldn’t have replied. but they replied to a post that said something, and they seemed to disagree with the post. so i also replied to their disagreement.
000
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
lol ok
000
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
the only way you and i can help to treat the global south better is to engage in politics though. which is what the original post said
100
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
no, it’s more like person A: *participating in society* person B: the real problem is that society sucks. cool but until you come up with a solution, were gonna keep participating in society.
110
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
you are expressing frustration at the fact that american politics has outsized effects on the global south, due to our immense privilege. but without suggesting a solution, i’m not sure what exactly you want to happen. you can only change politics by engaging in politics. sorry!
120
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
bsky.app/profile/mikh... i think the actual thing he said matters more than the emotional feeling he was trying to evoke.
100
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
“it’s objectively true” is genuinely arguable, since he said “pound for pound” “it needs to be said” is what everyone here disagrees with. i don’t think it is necessary, or even beneficial, for Piker to say this at the current moment.
100
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
once again, you are probably correct on what the guy was actually trying to say. i don’t know the guy. all i am saying is that a well intentioned reader can easily read this article and come out less informed than they came in. the article is bad. smart people sometimes write bad stuff. it’s ok
110
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
i am not mounting a defense of “rat fic” or yud, if that’s what you think i am doing. and in context, i do not feel led to think of machine learning, because of the database framing that we already talked about.
100
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
that word does not appear in either of the screenshots……..
110
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
“the chatbot might look into its training data to find instances…” “the chatbot was designed to do exactly…” nothing in those screenshots contradicts the idea “an ai agent can not do any things that are not explicitly laid out in its training data”
100
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
and once again, *even if that’s not what he meant*, it’s a coherent reading of the article he wrote,’if not a coherent reading of doctorow’s entire body of work.
100
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
3. i do not think that is a narrow critique whatsoever, as the main point of his article is to communicate the limitations of LLM capabilities. “LLMs can only do exactly what is in the training data” is wildly inaccurate.
200
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
1. agree to disagree 2. info about the author has nothing to do with the content of his educational materials. all his readers cannot be expected to have the context you have. we can discuss whether doctorow knows about LLMs, but that is a separate question from his article!
100
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
ok well, if you need to follow his career to understand the write up then perhaps it’s a bad write up. i hate to be a debate-bro but this is just a naked appeal to authority. basically the same kind of thinking that leads people to think musk is credible.
100
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
i also notice the retreat from your reply to will stancil, which indicates that anyone who thinks the article is misleading must not have read it.
100
fexonig.bsky.social @fexonig.bsky.social · 17/09/2026
having just looked him up, i am unsure why i am supposed to be overawed by a “Canadian-British blogger, journalist, and science fiction author”
200