Alessandro Del Rosso

Lifelong tech enthusiast (Commodore 64 → Apple Silicon). 12 yrs editor & partner at Punto Informatico. Since ChatGPT: hundreds of AI projects, apps & automation

[Alessandro Del Rosso @alexdelrosso]

GitHub Digest — 12 October 2026

ProxyBridge, InvokeAI, S&box, Netcatty, seomachine — proxy, generative AI, game engine, SSH and SEO content tools.

https://t.co/0m5e35ArwB

[Alessandro Del Rosso @alexdelrosso]

A cyberattack sent supermarket fridges in France into forced defrosting

Hackers set some Danfoss fridge controls to continuous defrosting at 50 degrees, with losses that could reach millions of euros.
https://t.co/tAl7feqloV

[Alessandro Del Rosso @alexdelrosso]

The Moon as a base: the US and China race to inhabit it

The lunar south pole, with its water ice in craters, is where NASA and China aim their first permanent bases.
https://t.co/H7nAPHGSID

[Alessandro Del Rosso @alexdelrosso]

Dalla SEO alla GEO nel 2026: Architetture Web, Microdati e Digital Provenance per gli Agenti AI
https://t.co/8624sXmqTc

[Alessandro Del Rosso @alexdelrosso]

Qualcomm: by 2028 AI firms want phones running 100-billion-parameter models

CEO Cristiano Amon says some leading AI companies are asking for phones able to run such models continuously on the device.
https://t.co/K1EJrOkdoK

[Alessandro Del Rosso @alexdelrosso]

GitHub Digest — 11 October 2026

Antibody, HelixSR, RAG_Techniques, Insomnia, garden-skills — AI agents, upscaling, RAG, macOS utilities and web libraries.
https://t.co/taNyG9gZPR

[Alessandro Del Rosso @alexdelrosso]

@ivanfioravanti @NTTLuke Ciao Ivan, sbaglio o la differenza di qualità tra fast (AR 8bit, 8 passi) e standard (AR 8bit, 32 passi) è quasi impercettibile?

[Alessandro Del Rosso @alexdelrosso]

Opus 5.5 vs Qwen 3.8 Flash Next... I love Qwen 3.8 Flash as a local model, it even runs on my modest 64GB of RAM. But against a frontier model like Opus? No contest :)

[Alessandro Del Rosso @alexdelrosso]

GitHub Digest — 10 October 2026

pm-skills, textual, FinderSearch, steamac, animate — AI agent skills, terminal frameworks, macOS and Linux tools.
https://t.co/kJKjIL9Wxd

[Alessandro Del Rosso @alexdelrosso]

I'd say if a free model that runs on almost any computer can generate this stuff in just a few minutes, the music composer's job is pretty much over. What do you guys think?

Model: YuE2-3B

[Alessandro Del Rosso @alexdelrosso]

AI companies prepare response plans for a major hacking attack

Executives at OpenAI and Anthropic reportedly plan responses if a catastrophic AI incident sparks public and political backlash.
https://t.co/JNgNQXYLbE

[Alessandro Del Rosso @alexdelrosso]

Style prompt:
"English, dance radio mix, progressive trance meets techno, sultry breathy sexy female vocal, driving four-on-the-floor kick, rolling offbeat bassline, hypnotic arpeggiated synths, euphoric supersaw build-ups, energetic club production, 130 BPM, A minor, 4/4"

Lyrics (verse / chorus / verse / chorus / bridge / chorus):
[Verse] Midnight pulse beneath my skin / Lights go low and we begin / Feel the bass inside your chest / Come a little closer, no rest
[Chorus] Move with me, move with me / Lose control and set it free / Hold me tight until the sun / We are never, never done
[Bridge] Breathe in, let it rise / Fire in your eyes / Drop it down, take me higher / Burn the night, set the sky on fire".

[Alessandro Del Rosso @alexdelrosso]

Before a recent post by @ivanfioravanti, I had never heard of YuE2-3B music model. I have to say, the quality really impressed me: we're genuinely at Suno levels, if not beyond!

The track below is a progressive trance x techno with a sultry female vocal. Just 3 minutes of processing on an M2 Ultra, running fully local. 8 steps were all it took.

The setup:
• Model: YuE2-3B by M-A-P, ported to Apple Silicon (MLX) in the mlx-Yue repo: https://t.co/iURKIONCBN
• Weights: vanch007/mlx-Yue2-3B + m-a-p/YuE2-Vae on Hugging Face
• AR 8-bit quantization, 8 midpoint steps ("fast" profile), 48 kHz stereo
• Driven by Claude Code through the repo's yue2-music skill

Prompt in the replies 👇

[Alessandro Del Rosso @alexdelrosso]

@ivanfioravanti Small!

[Alessandro Del Rosso @alexdelrosso]

@MiaAI_lab For my setup — a Mac Studio M2 Ultra with 64GB — the sushi + Qwen3.8.Flash 2bpw combination is still unbeaten!

[Ivan Fioravanti @ivanfioravanti]

Lux In Music experiment in progress with @NTTLuke
Suno killer is here: Yue2 3B of pure music power!
Here running on M5 Max

Model here: https://t.co/UH6FB8ftuO
Original Repo here: https://t.co/kzHNrzEveL

Lux In Music coming soon! Text to music, Studio, Stem, Edit and more. 🚀

[Alessandro Del Rosso @alexdelrosso]

@ivanfioravanti @NTTLuke I can't wait to get my hands on it! ;)

[Alessandro Del Rosso @alexdelrosso]

GitHub Digest — 09 Ottobre 2026

smolvm, summarize, Ouroboros, Chat2DB and OpenPencil — AI agents, summaries, databases and design.
https://t.co/vxYGoOjOtF

[Alessandro Del Rosso @alexdelrosso]

The accuracy is surprisingly good for a 2-bit quantized model. I left it working all night on medium-complexity projects and it always finished them. Obviously you can’t expect this model to generate the AAA games you see around today with Opus or Astra, but for less demanding tasks it’s a real workhorse!

[Alessandro Del Rosso @alexdelrosso]

@Beamsters1 I completely agree...
https://t.co/AIHK1hHoXd

[Alessandro Del Rosso @alexdelrosso]

Sushi 🍣 1.1.1 runs Qwen3.8-Flash 2bpw on a Mac Studio M2 Ultra with 64GB RAM: peaks of 500+ t/s prefill, 55 t/s decode at 262K context, in the Pi agent harness. The t/s counter is a custom Pi extension I built.
@Beamsters