0xcats (@0xCats) profile picture

#InfoSec & Hardware Hacker #SiliconChips #Linux #OSINT #BlueTeam #DigitalForensics Also on https://teapawt.net/t.co/J1bFoIxZkk https://teapawt.net/t.co/FOSb5sX8Dv maintainer

Namibia
Joined October 2016
Photo by 0xcats (@0xCats)Photo by 0xcats (@0xCats)Photo by 0xcats (@0xCats)Photo by 0xcats (@0xCats)Photo by 0xcats (@0xCats)Photo by 0xcats (@0xCats)Photo by 0xcats (@0xCats)Photo by 0xcats (@0xCats)Photo by 0xcats (@0xCats)Photo by 0xcats (@0xCats)
AI is amazing. I am extremely pro-AI 1. It has lowered the barrier of entry for programmers, resulting in hundreds upon hundreds of slop applications vulnerable to everything. This is job security. 2. AI influencers keep saying AI is going to destroy cybersecurity. This is good. AI influencers don't understand the size and scope of cybersecurity, they think it's just smashing a keyboard and making cat noises. This makes people less likely to enter our field, making us more valuable, making us more money. It's job security. Keep telling people cybersecurity is dead. 3. It's given us a new area of research: AI security 4. It's made task automation easier with slop Python scripts. In summary, cybersecurity is dead. DO NOT try to work in this field. It's all over. Cybersecurity has been solved!
107
338
28
4,651
152,938
Apparently, there is new hardware that can run LLMs much, much faster. It can easily give you over 15,000 tokens per second, which is stupidly fast. From what I understand, it's 'hard-coded,' meaning that you can't change the weights or anything; the chip itself can only run the model from the day it's manufactured, forever. Interesting concept.
227
157
81
2,766
288,361
0xcats retweeted
To PCB beginners: prepare to have your mind blown. To Experienced PCB designers: this software is very expensive. But I have personally used it, and it is amazing. This video is 11 years old. PCB AI Developers: This is the bar. If you cannot give this or better, you will likely not sell the most experience and the best designers. Furthermore, this company has deep pockets, and is developing its own AI tools within this eco system. Pay attention. Please enjoy.
52
81
9
1,220
117,470
found out that some SPI flash chips have a FACTORY MODE (presumably to make your production line run faster) which makes it write quicker at the cost of eating through cycle lifespan like crazy, incredible
15
32
617
38,888
This thing still on? Seems like it Anyways heres an update from me 0xcats.net/reverse%20enginee… Hp probably wouldn't want someone to know these numbers Until next time 😺
1
107
Beansed. More analysis required to make a definitive decision. But the fact that it's not immediately obvious speaks for itself.
1
1
277
For those wondering about those Ryzen 8000G Series APU's with Zen4c cores. They start life intending to be big boy cores. You can make a Zen4c core out of a Zen4(.1) core. Product binning and salvaging, isn't it fun. 😅
1
392
Friendly reminder to close twitter and do something else productive. You can do it 👍
1
1
4
300
Hey guys, A long-time member and friend of our Pixel Talk Discord is in a really bad place, and could use some assistance If you aren't able to donate, even sharing is a great help! gofund.me/b0e8b922
2
7
1
7
941
I really wonder what drove at @airsquare to request the domain administration details from clients for the purposes of server migration (CNAME record update) Yes, I confirmed this is legit. This really isn't acceptable procedure for the purpose of a CNAME record update. 😳
1
230
So I've double checked this and indeed my code is right. The Coppermine Celerons only have 512MB cacheable L2 cache. Likely as a result of their L2 being halved also some tag space is lost. But possibly this is also an arbitrary 'celeron' reduction.
Kinda made a odd discovery, not really sure if my code is right here or I'm being daft. But it appears that Coppermine (128K L2) Celerons (68a) only have 512MB cacheable address range.
1
1
313
Of course cache associativity was also reduced from 8-way on P3 to 4-way on Celerons.
140
Kinda made a odd discovery, not really sure if my code is right here or I'm being daft. But it appears that Coppermine (128K L2) Celerons (68a) only have 512MB cacheable address range.
1
1
3
636
On the other hand the old Mendocino Celerons have the (normal for Pentium 2/3) 4GiB cacheable address range.
1
174
This L2 tag size reduction reminds me of the early Pentium 2 Klamath chips with the 82459AB tag chip that were also limited to 512MB before the 82459AD tag + cache control chip on Deschutes raised it to 4GiB
1
157
Some late evening experiments I did. Cooked up CoreBoot BIOS for my Shuttle 440BX board to boot and run Windows 2000. Some tools are confused and think this supports UEFI already. Well not quite *yet*.
3
3
25
1,095
- In the first case Coreboot/SeaBIOS is kinda lacking advanced DMI table entries. - For ACPI I'm struggling my way through learning and writing asl (ACPI Scripting Language) its rather obtuse.
1
1
168
Some other things that need to be done includes merging the functions of the nvramcui utility for CMOS variables editing into the coreinfo utility which already shows lots of nice system details.
1
158
I discovered real-world @AMD #Zen4 speculative execution bug! #mitigations=off considered harmful! youtube.com/watch?v=1UnoBfw6…
1
2
186