adamame 🌾

@algoritmic 😭

Jira is Turing-Complete seriot.ch/computation/ji…

@omgidrawedit let us know when you're in Sydney! Vivid is on, heaps of cool light art installations all over

Gm. I’m in Australia now. Here’s a piece I don’t think I ever shared. Plants on the shelf. Felt quite Brisbane cafe vibe. Have great day!

@omgidrawedit I will happily accept your currency mate

Well all I can say is I sure do hope they accept Pokemon cards as currency in Australia.

@liamzebedee ✌️

You know for what it's worth, I find it really bizarre that if we are in the final moments before a singularity, no-one is just roaming the streets and filming what life is like before this seismic moment

@yjoonjang @capemox Thanks for the link, I hadn't seen these

@capemox @TheAdamEvans There are some recent papers trying to use saes rather than the mlm head - arxiv.org/abs/2603.13277 - arxiv.org/abs/2604.21511

Wow ettin does NOT like being fine-tuned for splade. Distilling didn't work at all, it's either representation collapse or not sparse enough trying a few different loss functions now, I've tried MarginMSE so far

@capemox creates a much smaller, fixed, (sparse) dim to work with downstream, basically turns it into an encoder for you might have better isotropic properties and behave better for your use case 🤷‍♂️ gonna read those @yjoonjang linked papers, seems like that's pretty much it

@TheAdamEvans zamn, tell me more basically, layer pruning qwen3.5-2b right? but why would qwen scope be useful, that's for interpretability right?

@liamzebedee @realGeorgeHotz when are you coming to Sydney for a visit It's just like San Diego but with fewer cars, a bigger harbour, better sushi, and way more sandstone

I still can't believe geohot was born on the fucking Gold Coast

increase thy information

If you think you are above labeling data you are not going to make it

@sama free people from having to work to survive

what problem do you most hope AI will solve in the future? maybe we can help!

@bubbleboi @redbrick87 bruh Sydney exists

@redbrick87 What else is there ?

Someone was trying to pitch me on this Australian chip startup and I was like seriously … Australian ? 😐

@bubbleboi at least we don't have super high tariffs with China 🤷‍♂️ Australia actually super capable brotha

@marek_galovic Love your work -- this is right at the edge of research and engineering that leads to real unlocks

SMVE solves both issues and enables online index updates (CRUD) + efficient metadata filtering while scaling to billions of documents with sub-100ms p99 latency and no loss in accuracy.

Cool talk on multi-vector retrieval, why it beats single vector methods, and why it's hard to use in practice. 🧵on what's next (1/n):

Information Retrieval is about making knowledge accessible. Late Interaction is the best way to do that today. But now that we have a new kind of users, it's time to zoom out so we can plan the future of retrieval. I gave a talk about this at @ir_tsukuba #slide=id.p" target="_blank" rel="nofollow noopener">docs.google.com/presentation/d…

@bclavie @xmlee97 @ir_tsukuba Love this walkthrough and context setting of IR throughout history and puts the different tradeoffs and research frontier in such clear terms. Great preso!

@alexanderrX_ 100% this and it's so hard to explain to people outside the industry can be thrilling but also quite intense

my job as an engineer has changed completely in 6 months.i used to pick up a backend ticket, write the code, add tests, open a pr, deal with comments, repeat. now i’m running 3-5 tasks async, steering an agent swarm of builders, reviewers and qa. it is not perfect. you still need someone competent at the wheel. but the job has changed. less pure coding, more architecture, product thinking, design taste and knowing what good looks like. some people say ai has made software engineering boring. for me, codex has been a 10x power up.

@joey_f6 less aha more HAHA amirite

It infuriates me when people say “aha moment”. Plz stop

@ekzhang1 Apropos 👏👏 I remember being so fascinated and inspired by that post

Homage to karpathy joining anthropic: made an in-browser (webgpu) version of the famous char-rnn "Unreasonable effectiveness of RNNs" demo, training on shakespeare :) ahh it's just as cool as when I was a student playing with keras in 2016