ＢＯＲＮ ＴＯ OOM CUDA IS A FUCK 鬼神 pkill -9 I am garbage collection 410,757,864,530 DEAD GRADIENTS
Joined May 2026
- Tweets1.1K
- Following586
- Followers125
- Likes8K
˚✧₊⁎ mage_owo ⁎⁺˚✧˖° retweeted
New paper out on symbolic local learning rules (to appear at NeurIPS, work w/ @StphTphsn1 and @Fabio_Ans)
We have a method to learn symbolic local learning rules that can be applied, in principle, to any task and any architecture
Link to the preprint: arxiv.org/abs/2611.12248
˚✧₊⁎ mage_owo ⁎⁺˚✧˖° retweeted
Stop waiting for permission or approval to become the person you want to be.
˚✧₊⁎ mage_owo ⁎⁺˚✧˖° retweeted
I think the data center backlash is the best thing to happen to AI. For years, cheap capital let builders ignore the communities and their neighbors. Now they must design for them. Quieter cooling, their own power, a share of the tax base. Show average-day Americans WHY they should trust this technology being built. This is a time of OPPORTUNITY.
˚✧₊⁎ mage_owo ⁎⁺˚✧˖° retweeted
A team of 1,000 Fable 5.2 agents fully formalized Fermat’s Last Theorem in 30 hours, translating the proof into a language a computer can check step by step.
With the same setup, Opus 5.5 completed only 26% in 18 hours, and barely made any more progress after 30 hours.
˚✧₊⁎ mage_owo ⁎⁺˚✧˖° retweeted
everyone roasting this deck
symbolica raised $31m for symbolic AI with no product, axiom got $64m at seed, taelin built HVM, plain code running on thousands of GPU cores, and asks $8m at $120m, most of you have shipped nothing close, relax
and I am not even talking about the startups that promised data efficiency and ended up writing GPU kernels (you know who), or the ones raising millions to train models stuck at loss 4 (you know who too) playing with nanogpt
omg no, I promise it wasn't what I meant :(
it is likely I'm being too stupid to even understand the ways on which I'm being stupid, but let me try to explain my reasoning
first, no, the deck is not a joke. it is what it is. we're raising up to $8m at $120m post-money cap, and our goal is to research a successor architecture to GPTs. is that a bad thing? I actually have no idea
now, on the humble part, I think the actual situation is the opposite... all big funds passed, and I think they see my work as a joke, so I decided to lock in until I have something worth of showing. this deck is a result of this emotional state, and not of any pride. is the ask high? perhaps I'm out of touch with the market, but I'd rather be over-protective and fail, than accidentally sell half of my company. in any case, I expected nobody would invest, which is why I put my own money. if HOC dies, at least I'll be in the same ship as my investors. it seems more fair this way
as to why I'm working on something ambitious (AGI) if I don't think I'm smart. well, it is because I think the approach will work. that's all there is to it. if you find a gold mine, you buy a shovel and dig. wouldn't you? sadly, shovels are quite expensive these days... regardless, the success of this project has less to do with my own capabilities, and more to with whether the approach scales or not, which is why I think it is worth trying
lastly, Bend's development is fully automated, so it isn't like there is much else I could do with my time anyway. AI is now capable of doing most of what I do. yet, rather than panicking like some mathematicians, I have a different perspective. I think this is an invitation to be more ambitious, and reach for things that were way beyond our own capabilities. thus, that's what I'll do.
(the cool part of reaching for the impossible is that, when things don't work, that's just another failure to your collection. you get used to it fast. yet, when they work, and occasionally they do, incredible things happen)
sorry for the text wall. I'm trying to write less, but I thought I had to explain myself. I apologize if my deck offended you, or if it was too dumb, or anything at all. tbh I don't even know what the actual criticism is, but I promise I mean no harm, I'm just doing what makes sense to me. please do not invest if you don't know what you're doing, and read the skeptical posts before you do.
I still don't know why my profile has much reach. I think people are entertained by my strangeness. in any case I'm glad to make your day a bit more entertaining, even if that's by laughing at my own failures. life is short.
I hope you all have a great day every day . . .
˚✧₊⁎ mage_owo ⁎⁺˚✧˖° retweeted
SOME THINGS AREN'T TAUGHT—THEY ARE REALIZED.
The confusion for us is we confuse life ( life as in nature) with a system which is not part of nature but a system created by humans to control and to govern us.
The system is designed for us to oparete in the level of the ego self. Now this creates a conflict with our true nature. We believe we are struggling and battling in life where in a true sense we are battling with the system that is controlling our experience within life.
We spend so much of our lives searching for answers outside ourselves, through people, knowledge, experiences, and the world around us.
But sometimes, the wisdom we seek begins with innerstanding ourselves.
True growth isn't always about learning something new. Sometimes, it's about seeing what was already there but we were too distracted to notice.
The more deeply you innerstand yourself, the more clearly you begin to see the world.
The question is,
Are we truly listening, or are we simply hearing?
˚✧₊⁎ mage_owo ⁎⁺˚✧˖° retweeted
are you sure he was not an ai safety non profit member of technical staff
˚✧₊⁎ mage_owo ⁎⁺˚✧˖° retweeted
I'm (ofc) super interested in studying systems that do NOT have attention anywhere (nor anything like it), yet that show signs of useful intelligence (like answering English questions). I'm collecting a list of these. Appreciate pointers. Could be anything
