Thinker, Software engineer, UITS Shooter.
Catania
Joined June 2007
- Tweets15.7K
- Following8.9K
- Followers12.3K
- Likes2.1K
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: gslf.it/th-ink
3
1
40
Gioele SL Fierro retweeted
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 (GitHub.com/unclebob) 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.
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
53
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.
1
29
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.
32
How much I learned from the latest UNIMATRIx benchmark runs
publish.obsidian.md/gslf/nem…
🤖 Made with AI
39
Gioele SL Fierro retweeted
Hyprland 0.57.0 is finally here, with tons of new stuff you don't wanna miss!
hypr.land/news/update57/
Gioele SL Fierro retweeted
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 retweeted
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.
block.xyz/dessa
Gioele SL Fierro retweeted
Turn of your Tailscale SSH, people 🫠 tailworm.com
‼️ Researcher Erica Windisch skipped coordinated vulnerability disclosure and published 11 unpatched Tailscale flaws.
In her lab, Tailscale SSH race conditions let one hacked device get root on other servers and add a rogue node.
Readers added context they thought people might want to know
The Tailscale SSH flaws detailed require a compromised but approved device on the tailnet that has been granted SSH access by policy. Tailscale has reviewed the report and stated they do not recommend disabling Tailscale SSH for everyone while they prepare patches. tailworm.com reddit.com/r/Tailscale/co…
AI benchmarking but with another approach. We don't even have a definition for intelligence, so why oversimplify everithing when AI is involved?
youtube.com/watch?v=Wca9ZPYo…
1
106
And finally the leaderboard for the HELL v1 benchmark is out. Got any other models you want to test? Just keep in mind our resources are damn limited . . .
32
The relationship between Wikipedia and LLMs is philosophically fascinating. Dwarfs and giants, trading places.
130
In Sweden, public employees can only use Copilot, or cloud AI services through an in-house condom.
Bullshits. If security is a real problem, you run local models. Otherwise, whatever. And what the hell is this magic contract with Microsoft that fixes every security issue?
127
Gioele SL Fierro retweeted
How on earth did this happen:
The Indian certificate authority (CA) eMudhra had one job (well, two):
1. Create secure, 384-bit private keys
2. Store these securely and do not leak them
Well, eMudhra chose private keys between 1 and 256 (the numbers!!) and called it a day. WTF
