Principal Consultant @ Brown Fine Security | IoT Security Researcher | Soli Deo Gloria
/dev/ttyUSB0
Based in United States
- Tweets 2,646
- Following 1,146
- Followers 11,800
- Likes 4,688
Is that from their API?
This is GLM 5.3 Flash hosted locally. You just have to push it and it will basically do it all.
1
1
103
This is insane. We cannot allow AI agents to go into your bank account and act independently without your authorization. But that is what’s happening right now.
Left unchecked, AI could endanger the entire global financial system — including your bank account.
Pause AI NOW!
Community note
Per Axios, detected AI agents access bank accounts with the customer’s permission. This violates most banks’ terms requiring human logins, which is why banks call it “unauthorized.”
Source: Axios (Oct. 22, 2026) — axios.com/2026/10/22/ai-
2
1
20
860
I'm noticing decode tok/s maxxing is less important for hacking tasks.
This is from a gray box hacking session consisting of:
- static binary RE
- root shell on device for dynamic work
- active network service testing
Notice that decode tok/s is not the bottleneck...
It's tool calling. Agent thinks about a problem, makes a tool call, then thinks about the result. rinse and repeat.
I think this is really good news for pentesters wanting a good local LLM setup. I've been getting solid real world results from my 2x sparks running GLM 5.3 Flash.
(btw this session found a crit in an IoT device)
11
1
44
2,151
Just gave GLM 5.3 Flash USB access to a cell module on this gas tank monitor and let it go crazy 😍
All Local on my 2 Sparks. No more giving my alpha to frontier labs. I'm getting around 55 tok/s with 1M context window. @MiaAI_lab recipe FTW.
GLM 5.3 Flash with its vision is also great at understanding pictures of PCBs and knowing what components are in view. Even can point you towards likely debug interfaces! Its a great hardware hacking assistant.
All I needed was a sketchy soldering job and a dream.
16
10
231
10,447
RTL-SDR. Same freq.
Just confused how you are so confident about something that isn't on your bench...
1
2
602
I bought an industrial radio controller made in China that transmits on unauthorized frequencies in the US. As it turns out, there are many of these knock off units being sold on ebay, Alibaba, etc.
In order to safely test the units, I used an RF enclosure to be sure that these unauthorized signals were not transmitted outside of my lab environment.
Using a USRP Software Defined Radio (SDR) I was able to observe that these radios transmit reserved for cell service and commercial aviation air-ground systems.
Additionally, I was able to perform a simple replay attack on the controller signals to successfully actuate the receiver.
19
74
899
138,103
I'm seeing at least 1.5x single stream decode improvement with TensorFold vs vLLM.
This is awesome and locks in GLM 5.3 Flash as my preferred model over DeepSeek v4/v4.1 Flash.
On my super unscientific benchmark (github.com/nmatt0/ihealth-sc…) I'm seeing it finish in 20mins on my sparks vs 15min on Opus 4.8
I've always said that Opus 4.8 level intelligence at home is my end game. it's basically there right now!
Run GLM 5.3 Flash EXL3 with TensorFold ⚡️
This is a completely new recipe that ushers a whole new level of performance for @NVIDIAAI 2x DGX Sparks.
Conservative default for stability:
- 1M context by default
- 2.7M KV cache pool (!)
- Yes, it's not a typo - 2.7M KV in just two Sparks
- 4 concurrent streams by default
Performance:
- 60 tok/s on prose, single stream.
- 108 tok/s on prose, 4 concurrent streams.
~1,950 prefill tok/s for most context lengths.
Stress-tested to handle a variety of workflows.
This is by far the BEST model to run if you have two DGX Sparks. Extremely smooth experience!
Expect further improvements!
Thanks to @ashxhart for developing such a powerful engine! TensorFold will be used in many of my upcoming recipes.
Get it here:
github.com/MiaAI-Lab/GLM-5.3…
3
1
36
11,176
I recently got a new hot air rework station (my old one died 🥲 )
Decided to kick tires on it by doing some chip-off firmware extraction and then some unpacking with Moria and analysis with Mithril.
Check it out: piped.video/Sr_ypTc9ZHo
2
4
66
3,205
I'm so glad that the government is stepping in and slowing the free exchange of good.
Gotta love a deadweight loss...
2
30
1,471
I downloaded the android app. If I read this right, if you turn on Bluetooth and don't explicitly turn it off then it will remain on...
1
1
20
LOL now we need to do the LLM version of this meme in case models are poisoned to not report back certain things :D
5
302
I've recently acquired 2x Nvidia DGX Sparks and have been kicking the tires on running local models for IoT security research tasks.
Check it out:
piped.video/DlTC6LfWmP8
14
29
186
18,575
Still going with Arch but I'm going to try wayland + hyprland instead of my usually Xorg + i3.
40
6
218
8,849
Wanna find IoT vulns before you ever get your hands on the device?
Download some firmware and lets go!
In this video we show off how moria and mithril work together to identify, unpack and mine IoT firmware for secrets, insecure configs and more!
Check out the tools at the links below!
moria: github.com/nmatt0/moria
mithril: github.com/nmatt0/mithril
5
49
322
15,323
"Claude, reball this eMMC BGA chip! Make no mistakes!" 🤣
There are a lot of things that LLMs can speed up, but sometimes on hardware pentests you only have one way to get the firmware: chip-off firmware extraction. Claude and Codex aren't going to be of much help for this task...
But what if you only have access to a single device? Ideally you would have more than one device when performing destructive testing but sometimes you don't have that luxury. So after removing the chip and dumping the firmware you need to get it back on in a functioning state.
This is where BGA reballing comes in.
My preferred method:
- BGA stencils
- solder paste
- hot air
Some people prefer solder balls but I find those difficult to place.
5
1
39
2,131
i have a degree in electrical engineering
i dont mess with hardware for the same reason i dont mess with linux
it’s a pain in the ass… but not any more!
i can talk to my ai harness on my omarchy linux machine out loud while i work…
and i gave it vision so it can see my circuits and give feedback…
and it can ssh into my controller and install/run code…
and it can ship prints to my 3d printer…
right here it is controlling a laser and taking test readings and telling me how to adjust the sensor in realtime…
it feels like magic!!!
1
1
30
1,958
We've got Opus 4.8 at home!
Running DeepSeek V4 Flash Vision-Exp on 2 x DGX Spark with OMP.
This is a mini-benchmark firmware analysis task I've been using to test out different models/harnesses.
Opus 4.8 didn't find anything in this firmware that DeepSeek didn't.
Thanks @MiaAI_lab
github.com/MiaAI-Lab/DeepSee…
13
23
318
26,147
2x Sparks running GLM 5.3 Flash!
I've got a mini-benchmark IoT pentest problem I've been giving to models/harnesses.
It's based off my favorite RE + crypto vid: piped.video/watch?v=hqau6n6e…
24 tok/s average and it solved the problem in 40 mins. (Opus 4.8 solves in ~15 mins) both used up about the same amount of tokens.
5
3
98
5,214
Me and the boys sailing through cyberspace with our newly issued letters of marque.
well shiver me timbers, they’re making letters of marque for cyberwarfare!
whitehouse.gov/presidential-…
7
17
214
11,853
