If this is true, it looks like Steve Jobs - Next playbook in action! 🍿
🚨 THIS IS ACTUALLY INSANE.
Jensen Huang is reportedly looking to raise $50 BILLION to launch a brand new semiconductor company.
He built Nvidia from a Denny’s booth into a $4T empire. The board says he’s not coming back. But with his relationships, he could plug into the entire AI ecosystem on day one.
Imagine the customers lining up before the first chip even ships...
Could Jensen really launch a real competitor to
$NVDA
, lock in $50B, and start taking market share from day one?
Hell yeah. If you don’t keep me here, someone else will.
1
7
Domenico Di Donna retweeted
Meet Allo! 👋
Your friendly neighborhood font, with variable weight and roundness, automatic optical sizing, true obliques, clever and stylish alternates, and loads more thoughtful features and details.
Free & open source: allo.joshpuckett.me
Domenico Di Donna retweeted
Same prompt, four Claude models, one Blender video each: "a Christmas tree that decorates itself" 🎄
Each one built everything in code with a single Claude Code agent in its own Herdr pane.
Fable 5.2 ⏱ 3 hr 58 min 💰 $41.58
Opus 5.5 ⏱ 4 hr 58 min 💰 $56.81
Sonnet 5.5 ⏱ 5 hr 6 min 💰 $20.52
Haiku 5.5 ⏱ 52 min 💰 $1.43
Notes:
- Haiku is the only one without sound: it couldn't find a free recording it was allowed to download, and didn't fake one. 🤷🏻♂️ Video in first comment.
- Video without model name is Fable 5.2. I added this detail later.
- I've used Claude Max subscription, costs is in the API usage scenario.
Now time to test some Open Models on the same prompt. 🔥
This is such an insane workflow. 5% Tripo3D, Blender, and Claude Opus 5.5.
The workflow in English, from Kamome Ashizawa's (葦沢かもめ) write-up:
note.com/ashizawakamome/n/na…
1. Give Claude 3 reference images and a short backstory
2. Harness: CLAUDE.md plus subagents for planning, modeling, rendering and visual checks
3. It plans, then builds in 13 stages, rendering each from the reference's angle to compare
4. No downloaded models, textures or HDRIs
5. For photoreal: light and color first, then materials, shapes, lived-in clutter
It even wrote a check that found 45 props clipping into things.
Video walkthrough: youtu.be/XbgbaOIBUHA
How the harness works: note.com/ashizawakamome/n/nc…
Blender's official MCP server: blender.org/lab/mcp-server/
teapaw.org/arafune/status/2118582…
Last week I said I’d work with smaller PRs.
Ihanged my mind,
1. Cognitive load:
Splitting everything and keeping track of what changes what gets messy, fast.
2. Disk space.
With Docker, every PR gets its own container. Tens of GB gone.
Fewer larger PRs, for now.
For 17 years I designed products sitting next to developers.
Two weeks at @cyberwave_com, and yesterday five of my PRs got merged into dev.
First lesson as a designer who ships code: small PRs, often.
My first ones were not small.
2
1
76
Something unexpected, I’m directing my agents and giving feedback after looking at the screenshots inside the PRs… sometime is faster than logginig in to local host, reach the page, and see the changes.
Not sure this is the best approach.
Anyone designing like that?
1
1
90
Wow, this is the end of an era…
NVIDIA announces CEO transition. Read the release: nvidianews.nvidia.com/news/n…
44
There are many breathing apps out there, with this one I tried to visualize the exercise a bit differently.
Try it for free
2
111
Domenico Di Donna retweeted
I asked fable 5.2 to invent a new musical form.
Fable 5.2 wrote a symphony that’s shaped like a training run.
The orchestra starts as random noise, learns the key, pulse, harmony and the melody, then plays music it was never shown.
For 17 years I designed products sitting next to developers.
Two weeks at @cyberwave_com, and yesterday five of my PRs got merged into dev.
First lesson as a designer who ships code: small PRs, often.
My first ones were not small.
🤯
Plot twist on Idol MLX: my tuned flash-attention kernel from M3 Ultra was 3x slower on M5, because it clearly not used the Neural Accelerators 🤯
Rewrote it with Metal 4 tensor ops: 50 steps went from 117s to 63s, at 31.5 TFLOPS 🔥.
Nanjing Niushoushan Culture Tourist Zone used as test here.
Idol now runs faster on my M5 Max MacBook than on my M3 Ultra 🚀 Can't imagine how fast this is on M5 Ultra.
Anyone willing to try? github.com/ivanfioravanti/id…
2
77
Interesting! I wonder if an agent like this could act as a QA layer for robotics sims, e.g. exploring scenarios in MuJoCo
We're open-sourcing Autopilot.
A self-improving agent that plays your game and finds bugs. Unity, Godot and Unreal.
→ 43% more accurate
→ Almost 2X faster
→ Beats Claude Code at playtesting
It’s Computer Use for games – closing the agentic loop and solving the most tedious part of game dev.
Stop being your agent’s tester. Free trial + GitHub below.
52
Domenico Di Donna retweeted
Replying to @kimmonismus
I do wonder how that will work… they seems all vague statements, without an actual scenarios described.
It’s also impossible (for me) to trust those statements coming from those who have ridiculous amount of money… the only scenario that UBI pictures in my mind are the ones of communism countries… and we all know how well those worked… or maybe will be the oligarchy of the AI owners.
Either way, I’m not optimistic on the outcome. (And I’m an optimist)
Then bending spoons acquires a software company that has too many employees, optimizes for revenue, because well isn’t charity, it’s a business, and everyone point fingers… sure all those must be entrepreneurs, they speak from experience.
(Yeah the tweet was out of touch, I agree, but that’s not my point)
Anthropic: $2 trillion valuation. $65 billion in revenue.
~5,000 employees.
Pinterest: 5,265 employees. Pictures for housewives.
DocuSign: 7,044 employees. Signs PDFs.
Zoom: 7,438 employees. Video calls.
4
185
















