I was in Budapest almost 20 years ago, and absolutely loved it. Sad to see the path it’s taken in recent years.

I actually do have a Mastodon account too, but unfortunately that hasn’t seen any activity in a long time.

I regularly run a script to unfollow people who haven't posted in 1-2 months, so please don't take it personal if I unfollowed you, just post more often!

This is why the Taalas chip is so interesting, recent open source models are already capable enough to handle a huge range of tasks. At some point the cost (and latency) of being at the frontier isn’t worth it.

I was surprised by the amount of littering, spitting, cigarette butts, drink driving, speeding, running red lights, etc when I first moved here.

Branchless programming.

sanixdk.xyz/blogs/the-we...

Modern DRAM is based on a brilliant design from IBM.

But, we're still paying for a latency penalty that's existed since the 60s!

In this video, I'm introducing my research project (Tailslayer) that immensely reduces p99.99 latency on traditional RAM!

I feel betrayed, how could MJ lie to me?

(As an aside, anyone (still) involved in crypto has like a 95% chance of being a scammer)

I came across BitDance (from ByteDance) arxiv.org/pdf/2602.14041 via this blog post: theadamcolton.github.io/small-llms-h...

I haven't fully digested it, but it looks like a fruitful direction - modelling images in an enormous binary token space and then using a diffusion head for decoding.

This probably won't shut him up, but this shows how wrong Gary Marcus was when he said it was impossible to generate an image of a horse riding an astronaut.

This was the character reference

Some pretty amazing reference-guided generation from Seedance 2.0 @coffeevectors.bsky.social

I wasn’t 100% sure until I saw the exact same comments repeated verbatim further down the chain, that was definitive for me.

1/n 🧵 Introducing Gaussian Wrapping — a principled framework for extracting high-quality meshes from 3DGS! 🚲
We recover thin structures, like bicycle spokes, where all prior methods fail.

Follow the thread for a brief overview and links!

Just meant I never really watched Star Trek so it's not something I've seen myself.

I've never seen it so I don't know if it's true, but if so, this seems *way* too prescient!

It’s like those non-degradable plastics that stick around for centuries, long after the person who consumed them has died.

Hah I haven’t actually seen that yet - what goes on there?

Reddit is so infested with bots, it’s becoming absurd. I’m seeing this so regularly - two accounts having conversations in entirely different languages (and if you scroll further down, the comments end up repeating themselves).

"Light Fossils" dinosaur light painting by artist Darren Pearson, known as Darius Twin, using a unique photography technique.

The artwork is created using long-exposure photography, where the camera shutter remains open while the artist moves handheld lights to "draw" the figures in real-time.

Today I spent some time optimizing clip masks. Previously these were quite sluggish when scrubbing through the timeline, very zippy now.

I like Canberra as a place (well, at least I did when I lived there 15 years ago) but I shudder to think what it would be like to raise kids surrounded by 100,000 people who've only ever worked in the APS.

I think another good rule is "don't believe anything coming from any company that's only a few months away from its IPO".

"AGI by 2026-2027" is the main one.

It would ring really hollow if they claim they've reached it but fail to put their money where their mouth is.

Not sure how you can claim AGI if it can't handle requests to refund $100 monthly subscription fees.

Until Anthropic deploys autonomous customer support with full authority for billing cancellation and refunds, take all their claims about AGI with a grain of salt.

CEOs of bulletproof glass companies literally stand behind their product while someone shoots at them. That’s real skin in the game.

Welcome to Jamrock.

GitHub - boksajak/MLPZen Contribute to boksajak/MLPZen development by creating an account on GitHub.

A gentle reminder my MLPZen project is up on GitHub. You get:
- MLP inference and training in standard HLSL, in-line anywhere in your shaders
- 2D and 3D Hash-grid
- Adam optimizer
- Dx12 sample app
- Link: github.com/boksajak/MLP...
Looking forward to how people will use this!

Seeing the step changes is amazing too. My son just got back from a week in China and his Mandarin probably doubled in that time.

Large scale video pre-training strikes again, this time for fully body avatar reproduction.

junxuan-li.github.io/lca/