Gioele SL Fierro

Thinker, Software engineer, UITS Shooter.

[Gioele SL Fierro @gslf_dev]

I’m fucking sick of hearing this bullshit, so I had to do something.

[Gioele SL Fierro @gslf_dev]

Modelli generativi che non generano. I costi si riducono drasticamente, e per alcune applicazioni specifiche possono essere anche più efficaci di un classificatore tradizionale o di un LLM.

https://t.co/qpvC12wxJs

[Gioele SL Fierro @gslf_dev]

I have a very specific opinion about markdown-based note-taking tools.

A PWA with real-time markdown rendering, cloud sync on Filen, a built-in static site generator, stupidly fast, and it runs on every platform.

Want to break it? Go ahead: https://t.co/Bhd3YWztlL

[Uncle Bob Martin @unclebobmartin]

Disengaging from syntax is not disengaging from engineering.

I’m working on a significant project at the moment. It’s in a pre-production state so far. More than a prototype, less than ready for real users.

I’ve been making multiple passes over the structure for each of the following concerns: Architecture, coverage, race & concurrency, security, API, etc. Each of these requires significant interrogation and review of verification procedures; but not review of the code.

At base are the /ub-mode (https://t.co/L3UhDJXshE) rules that keep the agents disciplined as far as coding structure and tests are concerned. But above that it’s me making sure that all the bases are covered well enough to keep my client safe from stupid sloppy mistakes.

I am the engineer. I am responsible. The agents are just my tools. They are power tools, to be sure. But power tools in the hands of novices result in missing fingers. So I am very careful.

[Gioele SL Fierro @gslf_dev]

Nemotron 3.5 Super is not going into main branch of inferenZ. The port was validated layer by layer but the performances are not enough.

Decode: 3.95 tokens/s against Qwen's 7.53
Coding: 27/40 against 33/40
Weight reads per token: ~10 GB against ~6 GB

[Gioele SL Fierro @gslf_dev]

Tried Poolside's Laguna S 2.1 (118B MoE) in inferenZ, my C LLM engine for a 62 GB laptop: 6 tok/s and 26/40 on my coding eval vs Qwen3.8's 7.5 and 33/40, with 17-minute answers on average. Not merging. Bonus: tuning the kernels for it made Qwen 12% faster.

[Gioele SL Fierro @gslf_dev]

Tried NVIDIA Nemotron 3.5 Super (120B, Mamba-2 hybrid) in inferenZ, my C engine for 120B MoE models on a 62 GB laptop. NVFP4 weights copied bit-exact (KL 0.003). But 12B active = 10 GB read per token: 4 tok/s and 27/40 on my coding eval vs Qwen3.8's 7.5 and 33/40. Not merging.

[Gioele SL Fierro @gslf_dev]

How much I learned from the latest UNIMATRIx benchmark runs

https://t.co/Fwr9OkDyFc

[Hyprland @hyprwm]

Hyprland 0.57.0 is finally here, with tons of new stuff you don't wanna miss!

https://t.co/84vzB4ADbe

[Gioele SL Fierro @gslf_dev]

@thsottiaux 🤔

[Gioele SL Fierro @gslf_dev]

My @X feed is full of useless shit right now.
Why?

[Gioele SL Fierro @gslf_dev]

@Ionkosm Do you have a definitive definition for AGI, and reliable tools to evaluate it?

[Gioele SL Fierro @gslf_dev]

@NewsFromGoogle @GoogleDeepMind Minutes? Hours? Days? Weeks? Months?
How long for mere mortals?

[News from Google @NewsFromGoogle]

Today we’re introducing Gemini 4 Carbon — our newest frontier model from @GoogleDeepMind.

Carbon is our strongest model with state-of-the-art cybersecurity and long horizon capabilities. Built for sustained agentic workflows, it delivers improved performance across domains — with greater alignment, transparency, and follow-through to respond to high-pressure tasks.

We’re rolling it out starting this week to cyber defenders through our Fairwind Program, with expanded access as soon as possible.

[Gioele SL Fierro @gslf_dev]

@PrismML Bonsai is one of the new entries of UNIMATRIx Leaderboard

[Gioele SL Fierro @gslf_dev]

@Recursive_SI Flint is one of the new entries of UNIMATRIx Leaderboard

[antirez @antirez]

Credits where credits is due: kudos to @OpenAI in the context of the new scientific releases, for putting the scientists that worked in a given field for a long time at the center of the process of understanding and announcing the new discoveries.

[jack @jack]

We believe voice is the most natural way to interact with intelligence, so we created Dessa, the most authentic full-duplex speech model ever built. Talking with it feels like talking with a person…full of emotion, nuance, pauses, and a natural cadence.

https://t.co/r0qFLdvd6c

[Gioele SL Fierro @gslf_dev]

[ p ]

promezio

[Gioele SL Fierro @gslf_dev]

@OpenAI 6 Luna enter in the leaderboard.