iiinova — where your digital self keeps evolving

knowing the right time to be goofy is an underappreciated skill as well, people often equate unseriousness with disrespect and most of the time the only thing separating the two is timing

020

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 16h

i blame el niño, scientists did warn it'd impact where whales live

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 16h

me and my demons (externaldns)

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 05/10/2026

interesting, i will try deploying on my cluster. does it have a kustomize/helm?

000

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 05/10/2026

seems like a very vertical workload, given the state is an external object storage what stops you from serving it horizontally in an auto-scaling environment instead? wouldn't it be more effective?

120

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 04/10/2026

am i the only one that finds odd how every factor was against bolsonaro's candidacy until the u.s. started to focus on brazil?

050

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 04/10/2026

most of the weird phrasing observed recently is a symptom of the watermarking process, not an inherent property that frontier models developed after crossing some threshold of intelligence.

030

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 04/10/2026

given how ai still sucks at medicine i would believe replacing all coding with llms is not a good idea for critical systems, one of the core uses of rust as a language. the technology just isn't ready.

030

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 30/09/2026

this. this is the perfect answer

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 30/09/2026

started reading cisco's unlanguage and it's one of the most insane things i've ever read. definitely recommend if you are into horror and plots so dense they make you re-read the same page multiple times, every page of the book

000

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 30/09/2026

it is the same move anthropic did from fable 5.1 to opus 5.5, cheaper price for similar benchmarks (in theory), in practice it seriously under-performs opus for only 50% the price, not worth it imho

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 28/09/2026

they grow up so fast (╥﹏╥)

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 15/09/2026

true. it was mostly pods that should have run their course, completed their tasks and died in isolation. kinda crazy they didn't have an operator checking for that behavior btw

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 15/09/2026

scary but not surprising, anyone who knows how model inference works also are painfully aware of how transparent and leak-prone this data ultimately is, especially if they are hiring contractors to look into these conversations. what is more frightening is that this is totally legal in the U.S.

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 14/09/2026

agora o domínio do meu site finalmente redireciona pro lugar correto, não sei pq eu não consertei antes

000

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 14/09/2026

that's cool, wish more communities would start doing this

000

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 14/09/2026

their google cloud bills must be crazy

000

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 14/09/2026

to pensando em voltar a postar aqui

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 03/09/2026

probably the most egregious a researcher can do, not only it taints your work but also your whole network

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 02/09/2026

or you could make your own, the worst one, because it makes you hate yourself

000

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 02/09/2026

me when i open an issue asking for a new uselsss feature nobody asked for, tag the maintainer to speed things up, and they close it without saying why

030

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 02/09/2026

have you denounced your local twitch streamer today??

000

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 01/09/2026

literally so happy she became :3 emoji

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

my point is: if you want to llms to represent medical scientific rigor they shouldn't be bound by pattern-matching but evidence and mechanistical reasoning. it also answers your question: regardless of the framing it always arrive at the same answer, something experts can't in medicine

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

the reason why medicine hasn't is the same one that gave us vaccine skepticals: arrogance and mistrust of science. homepathy is recognised as a valid treatment here by the CFM, STAR*D proved to us SSRIs are worse than placebo and YET meta-analysis keeps confirming it works (barely). why is it that?

200

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

you are treating an expert opinion like it is a directive when it is most of the time an educated guess. the real answer to ur question is not who says 'a b c' is correct, but rather, how do u make sure there ain't more letters you don't know about? if theres a field that should ask ts is medicine

200

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

ironically, thats exactly the reason it was created. intelligence agencies are full of people who are experts, varying political biases and who were correct most of the time but had little to no predictive power when against something they didnt comprehend, a metric ACH tracks in real time

200

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

this is especially the use-case for AIs: it allows you to use zero-shot opinionated dialogues, but instead of a direct comparison between two models, you build a matrix. the signal doesnt come from where they agree with each other, but rather from when they dont even if biased towards it.

100

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

wait, she's a head of a democrat think tank? they might become republicans at this point istg

2260

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

it is their mid-term strategy, all the way through. any liberal hating on hasan is much worst than those who refused for kamala for example, they are actually sabotaging the first momentum democrats ever had in recent years out of personal grudge against him or financial interest

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

see? should've started with that but i appreciate it nonetheless, will read it and comment back

020

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

what did i say that was factually incorrect? you keep saying i dont know anything but refuse to say what i said that is wrong. like if im wrong id like to learn, differently from your attitude

100

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

i am afraid that's not the case. the introduction of multi-modal storage and retrieval, vlms, vlns and vlas allows llms to handle input like audio, video and derive something from them and trigger actions independently. if you would like to read further: arxiv.org/html/2311.07...

100

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

yea you absolutely schooled me there, surely i have no idea what im saying so dont bother reading and be rude somewhere else

100

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

medicine is a field that can be enhanced by methodologies used in intelligence operations: both work with high stakes uncertainty with progressive evidence discovery. using something like an ACH framework to a difficult diagnosis has been consistently a good way to identify what is missing

100

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 31/08/2026

yes, they are being controlled but there's not much of a gap on how to automate them in practical terms once they figure out how to place something like a llm and run it locally

100

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 30/08/2026

a biped robot just beat usain bolt's record, can compete in fights, etc. dont fall for musk's show of incompetency, there's real engineers working on a weapons race and the first clients will be billionares, for sure. once they figure out efficient local inference on robots we are kinda screwed

510

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 30/08/2026

ai is a broad term, what you described comes from training and inference, which is not a model by itself but rather the consequences of creating a model. when i refer to "ai" is a problem i mean that even if they were nuclear powered and waste-free, wouldn't make it less threatening

110

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 30/08/2026

this was true initially but things have advanced to some scary degree lately, frontier models are genuinely on a league on its own and if the same progress pace continues, things will change radically. in the security world they wrecking havoc even, and i consider them to be the canary in the mine

000

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 30/08/2026

because things like luigi mangione would be way more difficult with something like private robot armies escorting ceos, which is now a flourishing industry. they are literally gearing up to make others obsolete, as peter thiel said, he doesn't believe humanity as a whole should endure

1200

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 30/08/2026

the problem is not ai itself but what it represents. in a way robotics/ai enable billionaires to be able to replace and acquire cheap labor, the last bastion of power the working class had. it is their wet dream not having to pay a single cent more and automate everything

3341

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 29/08/2026

im starting a conspiracy theory to see if americans can learn about energy infrastructure unwilling to do so naturally, attracted by their americans sixth sense for psyops bs and fall for it hook, line and sinker

000

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 29/08/2026

se vc e 500 mil pessoas nunca mais mais tomarem banho e lavarem as mãos na sua vida da 1 dia de agua usado pela fazendas de gado mais fracas do rio de grande do sul

010

󠁧󠁢󠁥󠁮󠁧󠁿 @tou.moe · 29/08/2026

"are liberals drinking too much water?" - oped totally not paid by anyone important, mysteriously front page nytimes

000