Physical/infrastructural security researcher and penetration tester Cybersecurity management

Brno
Pred 20 rokmi mohli číňania hovoriť niečo ako "Najlepšie auto z Číny." Teraz my hovoríme "najlepší otvorený model z Európy", pretože tie najlepšie autá aj otvorené modely sú… z Číny.
Meet Mistral Large 4, aka Le Chonk. • 1T parameters, natively multimodal. 49B active. It is the best open weights model from US or Europe on aggregated benchmarks. • State-of-the-art on critical workloads, including cyber defense, manufacturing and finance and it surpasses closed frontier models on visual grounding. • Forged in Europe end-to-end and is deployable from Europe via our own Mistral Cloud infrastructure. • Available to all via API today. Working with cybersecurity partners privately. Open weights release end of October.
2
Marek Šanta retweeted
That’s why we had to cut RAM in half for Optimus lol.😂
Replying to @Bqlsj2023
“This is why RAM is so expensive” 🤣🤣
62
136
918
119,882
toto si odložím skôr ako ma diskvalifikujú 👀
2
Hey Inference providers, @vast_ai and others... Since you don't allow SYS_PTRACE, we can't run qwen3.8-exp models efficiently. Please fix that...
6
Marek Šanta retweeted
41
328
5,351
512,081
When @deepseek_ai Flash 4.1 is finally satisfied after the 12th round of code review
6
Marek Šanta retweeted
They didn’t bridge to $ZEC. Bitget already had ZEC sitting there. Attackers yoinked ~18.9k of it on the 24th to t1WgMdtND8NF7NDUuYmq8MpMj1NTCXkMDVG. That’s inventory theft, not some genius rotation. What you’re watching now is them stuffing ~2.7k of that same pile into Ironwood. Hide the later hops. Doesn’t wipe the deposit. When it pops back out to a t-addr the size and timing still line up, exchanges still treat it like dirty ZEC. Normal crew play is swap the loot into $XMR. No public history, nothing to blacklist. ZEC still has a transparent door. That’s why this looks like a speed hide, not the finish line.
Hory shiet... it seems the IQ of criminals is rising. They’re using Zcash’s shielded pool now. Now that they understand Zcash better, maybe they’re afraid of obfuscation-based, guessable privacy and finally want cryptographic privacy.
10
3
35
10,929
Komunálne voľby v ČR už sú klasikou komédiou...
Koukám, že v Praze se máte dobře :)
33
zcash:native's Ironwood Formal Proof is nice for the post-NU6.3 circuit. The problem is the "simplification to only one shielded pool" So Orchard<->Ironwood supply integrity is NOT covered by the proof 👀
13
So... bitcoin:native is so far perfectly copying the previous cycle, as defined by BTC maxis... Zoomed out in the next post:
1
28
This is not my call or prediction, just a playful graph drawing. In general, I deem BTC to be obsolete. monero:native is much more utilitarian, and without utilitarianism, you don't have true value. @wilderko, your opinion on this would be the most relevant to me...
21
Marek Šanta retweeted
Replying to @cheatyyyy
Ok. We entered slop stealth models. What a time to be alive
2
50
2,633
Our "Emil" server now runs two models. Contains an RTX 3090 with a cool story + Radeon AI R9700 from @tunasec_cz (thx!) Strict zero-data-retention policy (too lazy to log anything anyway) and maximum HarmBench score.
1
65
Marek Šanta retweeted
* 107$ Oil * 5.00% US 10Y * S&P500 2.6% from ATH One of you mf'ers is mispriced

ALT the simpsons GIF

71
165
3,736
175,638
Another day, another reason for $XMR ?
If you told me in 2011 that The Clarity Act would fail I would have asked you why the fuck you're looking to the government for permission to use Bitcoin.
1
56
🤖
🚀 Introducing DeepSeek-V4.1-Flash: smarter, faster, more efficient. 🔹 Introducing the smallest model in our new architecture family, with native visual understanding. 🔹 Designed for greater capability, faster inference, higher throughput, and scaling to larger models. 1/6
1
16
Marek Šanta retweeted
me opening X nowadays 😭
257
1,282
11,465
249,766
Marek Šanta retweeted
"just answer yes or no"
143
580
9,399
557,944
👀
The next Qwen wave is coming
1
31
UD-Q4_K_XL on RTX 3090 is incredible!
We’re releasing new Qwen3.8-27B GGUFs with 10% higher accuracy. Unsloth Dynamic V3 outperforms others by >10% on Div-300, KLD & more benchmarks. We also release 1-bit quants that retain 77% accuracy. Run on 8GB RAM. Blog: unsloth.ai/docs/basics/dynam… GGUF: huggingface.co/unsloth/Qwen3…
2
42