One annoyance abou Dart is that JIT/VM compile & startup is pretty slow. There's still no way to precompile parts of an application or library (e.g. if app A depends on package B, package B is always going to get parsed/compiled whenever I run app A in JIT mode, even if nothing's changed). BUT...

I wish I had 1/10th the confidence as a guy from Bengaluru doing cold outreach on LinkedIn.

America. (Funnily enough, applies equally to Singapore. No layoff protection, or even a legally-mandated redundancy/retrenchment payout)

I'm surprised by how many people seem to think they can build a financially viable business from an agent harness. This seems to be the todo app of the AI era - low stakes, easy to implement, but hundreds (if not thousands) of competitors who do the same thing equally well.

Ahead of the next Thermion release, I've been smoothing out the pipeline for headless (Dart-only) rendering. www.youtube.com/watch?v=qV82...

Headless 3D rendering with Dart

YouTube video by Nick Fisher

youtube.com

Dario, dressed in a grey Stasi tunic, in the interrogation room flashing a harsh prison light: "Please understand, we have many methods that would...prompt you to be forthcoming".

Sung Kim@sungkim.bsky.social · 4d ago

LOL. job-boards.greenhouse.io/anthropic/jo...

AMD is acquiring Taalas. I can't overstate how exciting the next wave of hardware is going to be. This is the 90s 3D accelerator race all over again!

Having a blast at #QuakeCon so far!

It's awesome to see 3/4 of the original id team at QuakeCon, this probably hasn't happened since the 90s.

Fabien Sanglard@fabinou.bsky.social · 4d ago

Growing up in the 80s in France, the USA was another planet I was unlikely to ever get a chance to visit. Sitting across Tom Hall felt surreal!

Now that I've finished moving house, I put together a quick video on my recent Unreal Engine plugin for exporting MetaHumans with ARKit blendshapes. youtu.be/JyZ4_GUCOno

Export MetaHumans from Unreal Engine with ARKit Blendshapes

YouTube video by Nick Fisher

youtu.be

How it started: How it's going:

DeepSeek just posted a notice that they're going to be raising prices "significantly" in the near future. The final V4-Flash is a noticeable improvement over the preview version, but if prices increase too far, it's going to much less attractive.

Has anyone told him that he really looks like Groucho Marx wearing those glasses.

Ruo Shui@ruoshuiresearch.bsky.social · 5d ago

someone wants attention

I've just started experimenting with Hermes Agent via Signal with voice messages and I gotta say...I may never type again. This is amazing.

Took my son to pre-school on a "family" sharebike (normal bike + child seat on the back), which was $4 for 30 minutes (compared to regular bike which is $1). How does that compare to elsewhere overseas? It feels very expensive, particularly when a round-trip taxi probably would have cost me $10.

Me, talking to my agents.

My Keychron mouse has started to fail smack-bang on the end of the 12 month warranty period (LMB is intermittently failing to register, so it stops registering mousedown halfway through a click/drag which is infuriating when you're trying to drag things around the screen).

I just want to repeat my previous sentiments that this work (and open weight model releases like DeepSeek) is vital.

antirez@antirez.bsky.social · last wk.

DwarfStar branchk "ds4f-mxfp4" now can run the lossless MXFP4 DeepSeek v4 Flash GGUF I published on my Hugging Face account. It rocks even with SSD streaming in 128GB systems at > 20 t/s in case you want to try the *actual* DS4F weights released without any quantization.

So many new features in vk_gltf_renderer (github.com/nvpro-sample...) since my last post, I'll drop a few highlights. - glTF 2.1 External Assets - Releases: compiled version of the sample - PBR: better volume scattering, retroreflection - Optimizations: material compile defines - Opacity Micromesh

Looks like the non-preview DeepSeek V4 Flash is now live on the API! api-docs.deepseek.com/updates/

Change Log | DeepSeek API Docs

Date: 2026-07-31

api-docs.deepseek.com

I’m happy to point the finger at the current US administration for a lot of things, but I won’t dogpile on this because it’s exactly the type of thing I would have done on a bet as a young IT guy. www.bbc.com/news/world-u...

US government lists fictional nation Wakanda as trade partner

Officials say it was an error to add the fictional nation from Black Panther to the list.

bbc.com

Aaaand another one. Seems like it's the real deal. github.com/PauliusOS/pa...

Nick Fisher@hydroxide.dev · 2w ago

Another one here. Performance is choppy and the built-in threejs toon shader does a lot of the heavy lifting, but the coherence of the world is blowing me away. They're also claiming Opus 5, no assets, all drawn from primitives. This is a real step change from last month. github.com/Kenton-GMI/s...

Another one here. Performance is choppy and the built-in threejs toon shader does a lot of the heavy lifting, but the coherence of the world is blowing me away. They're also claiming Opus 5, no assets, all drawn from primitives. This is a real step change from last month. github.com/Kenton-GMI/s...

Nick Fisher@hydroxide.dev · 2w ago

This person is claiming that Opus 5 made this in threejs on its own (no external assets, everything from scratch). If it's true...colour me impressed. Actual deployed link on itch is below www.reddit.com/r/singularit... heromouteut-ai.itch.io/walking-simu...

Ugh.

It’s been a long day in the electron mines, I’m off to bed. Unfortunately we’re moving house on the weekend so I’ll be dealing with that saga. I need to take some time to recharge soon, I’m running on fumes.

This person is claiming that Opus 5 made this in threejs on its own (no external assets, everything from scratch). If it's true...colour me impressed. Actual deployed link on itch is below www.reddit.com/r/singularit... heromouteut-ai.itch.io/walking-simu...

Opening up this one to the audience - any recommendations or suggestions for affiliate program platforms? Do I even need to bother - this feels like something I could implement myself fairly easily?

This is who runs this account.

GPT-5.3-Codex-Spark is the peanut dust of models, I have no idea why you would bother.

LATO factors mesh generation into a vertex flow followed by connectivity. In other words, 3D generation with halfway decent topology! lohhhha.github.io/LATO.2/