Principal AI Engineer & Architect. Building production AI. Open weights + AI economics. Building Omarchy MX Mac. [email protected]
Australia
Joined April 2009
- Tweets592
- Following995
- Followers2.1K
- Likes7.2K
Next mission in-flight:
Make DeepSeek 4.1 Flash 1017 usable for agents using Strata on the 3 3090s.
Reports incoming.
Marcelo Alcantara retweeted
If you're using @Tailscale, I highly recommend creating a network agent in hermes or @bot and asking it to lock it down. They have some amazing security features that are not enabled out of the box that I think our must haves.
Cred to @NetworkChuck for throwing that bone out months ago. My network agent is named Chuck 😂
Marcelo Alcantara retweeted
‼️ Tailscale SSH Flaws Could Let Hackers Gain Root Access Across Servers
Details: cybersecuritynews.com/tailsc…
A new security report warns that Tailscale SSH has flaws that could let a hacker move from one compromised device to root access on other servers. Tailscale SSH lets approved devices log in to each other over a private network. The report lists 11 findings in five groups.
They cover Tailscale’s SSH server, its serve and app connector features, the DERP relay client, and headscale, an open-source server that manages Tailscale networks.
The researcher tested Tailscale 1.104.2 and headscale 0.29.5 between October 22 and October 24, 2026. The most serious group involves Tailscale SSH. Its SSH server runs inside tailscaled, a background program with full root privileges on the machine.
#cybersecuritynews
Readers added context they thought people might want to know
The flaws require an already-approved device in the Tailscale network with SSH rules permitting access/forwarding; they enable post-compromise lateral movement and escalation inside the tailnet, not unauthenticated remote access from outside.
tailworm.com
brandefense.io/security-news/…
tailscale.com/security-bulle…
x.com/mmaunder/statu…
I benchmarked the RTX 3090 unlock mod on an ASUS Strix, stock vs unlocked vs stock, using random selected models I have been using this weekend.
Overall +64% BF16 TFLOPS
+49% LLM prompt processing (Qwen 27B)
35% faster music synthesis with YuE2
Decode and Wan 2.2 video: no change
No temp changes and identical outputs
This mod is a keeper. Will report back if I find any "surprises" with my workloads.
This was the craziest 2 weaks in mass scale security breaches.
Everybody getting hacked left and right.
I went on a side mission yesterday and put my Omarchy GPU server to work in something fun about it.
I hope it cheers your weekend start.
#pleasedonthackme
The new 3090/4090 GPU unlocking mod is great news.
But at the same time it's so disappointing to know we have a hardware that is more capable and was tuned down on the first place.
We need so much more competition on this space.
@fireship_dev would love your feedback. The horses may want to sign up to your horse tinder.
Marcelo Alcantara retweeted
Omarchy QA, using nightly builds, is now kicking off every morning at 8am
granted there is only one test, but that one test is beautiful :)
Should have 50 tests here soon
Thanks for the gifts Joshua!
I need to try this out.
I also love what they are doing.
Looking forward to the DeepSeek inference silicon.
Some great progress on the M3 GPU support.
M3 Max GPU is on, all 40 cores working with our aurora kernel.
Thanks to @ewninjaofficial and @SiliconForested for collaborating with me on this.
Next up: I’m joining @JoshuaSWarren to push Local AI on M3 Max
@dhh @OmarchyMac @AuroraSilicon @omarchy
AI has been a revolution of my homelab management. I don’t know how I could manage before without it.
I gave @grokbot my UniFi Dream Machine API for 1 night... and it tuned the sh*t out of it.
408 stale IPS rules, gone.
41 ghost clients, gone. 144 down to 103.
an old port forward I forgot about, gone.
mesh backhaul went from about 450 Mbps to as high as 1170.
my VPN Wi-Fi moved to WireGuard and doubled, about 200 to 400 Mbps.
It also caught my nightly backup silently skipping 14 days in a month. Now it keeps 30 days and yells if one fails.
every change backed up first, done 1 at a time, verified.
When did you last actually look inside your router?
Marcelo Alcantara retweeted
Linus is based
"[Linux] is *NOT* some kind of "social warrior" project, never has been,
and never will be." — Linus lore.kernel.org/linux-media/…