Casperinous (@Casperinous) header
Casperinous (@Casperinous) profile picture
Joined February 2013
Photo by Casperinous (@Casperinous)Photo by Casperinous (@Casperinous)Photo by Casperinous (@Casperinous)Photo by Casperinous (@Casperinous)Photo by Casperinous (@Casperinous)Photo by Casperinous (@Casperinous)Photo by Casperinous (@Casperinous)Photo by Casperinous (@Casperinous)Photo by Casperinous (@Casperinous)Photo by Casperinous (@Casperinous)
Casperinous retweeted
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
1,936
5,853
4,577
46,357
5,523,269
Casperinous retweeted
the cleverer among you can think of ways to use this, a trivial one is to lie to decompilers and lifters.
2
2
15
380
Casperinous retweeted
One of the best explanations on CPU branching ocw.mit.edu/courses/6-004-co…
1
45
2
469
28,832
Casperinous retweeted
There’s 100% a kind of mania you can get into with coding agents, where it feels like you’re getting a lot done and working all the time but nothing is really happening. I think it's often a form of avoidance of something else in life.
810
850
531
16,984
1,409,841
Casperinous retweeted
#VoltTyphoon operational network for 0-day exploitation uncovered. Very insightful research from the people @lumentechco . Kudos for using the insights on this network over a 3 year timespan to inform victims/targets before major impact. lumen.com/blog/en-us/silent-…
4
16
1,272
Casperinous retweeted
In an international law enforcement operation 👮, the US Department of Justice 🇺🇸, in cooperation with private sector partners, have taken down notorious bulletproof hosting provider Virtualine 💥 ➡️ justice.gov/usao-edva/pr/bul… Operating as AS202412 (Omegatech LTD), Virtualine received upstream connectivity from German 🇩🇪 providers aurologic GmbH (AS30823) and Pfcloud UG (AS51396), as well as French 🇫🇷 PEERXY NETWORKS SAS (AS213382), as bgp .tools shows. This network connectivity is now gone, currently making it impossible for Omegatech to operate 🛑 Omegatech announced 24x /24 IPv4 CIDRs, totaling 6,144 IPs. Out of these, hundreds were seen daily participating in spamming, internet-wide scanning, and botnet C2 hosting ⤵️ 🌐 URLhaus tracked over 10,000 malware distribution sites at Omegatech: urlhaus.abuse.ch/asn/202412 📡 ThreatFox logged 2,000+ community-shared IOCs tied directly to Omegatech: threatfox.abuse.ch/asn/20241… The hoster primarily served malicious infrastructure associated with RATs 🖥️ (e.g., PureRAT and OverlordRAT), rogue ScreenConnect servers 🔌, and social engineering malware campaigns like ClickFix and ErrTraffic 🖱️. While internet abuse for sure can happen, Virtualine was not reacting on abuse reports at all and advertised their "bulletproof hosting services" on the darkweb, as investigation shows 🕵️‍♀️ The takedown of Omegatech is welcome news. However, its long-term impact remains uncertain. Bulletproof hosters commonly re-brand and resume operations under new ASNs quickly 🔃 Furthermore, Omegatech is just the tip of the iceberg. Other active bulletproof hosters include: ➡️ AS219502 (STORMCLOUD) ➡️ AS207043 (DEDIK-IO) ➡️ AS214351 (FEMOIT) What do they all have in common? According to bgp .tools, they all receive upstream connectivity from German provider aurologic GmbH (AS30823) as well 🔍👀 If you don't want to worry about the threats posed by bulletproof hosters, just block any CIDRs and ASNs listed on @spamhaus 's Don't Router Or Peer List (DROP): 🛑 spamhaus.org/blocklists/do-n…
3
14
1
42
4,480
This is absolutely nuts and only hitting German papers so far: leaked emails show that seven politicians from Germany's far-right AfD travelled to Syria to meet with Assad regime officials... facilitated by Hezbollah AND Russian intelligence. handelsblatt.com/politik/deu…
47
1,639
89
4,448
88,243
Casperinous retweeted
Another day - another Stealer: MopsStealer 🐶 Active botnet C2s, hosted at Podaon SIA 🇳🇱: 📡 mops .to 📡 45.91.202.47:408 📡 45.91.202.47:1488 📡 45.91.202.146:408 📡 45.91.202.146:1488 Malware samples ⤵️ bazaar.abuse.ch/browse/signa… IOCs ⤵️ threatfox.abuse.ch/browse/ma…
1
8
32
3,223
Casperinous retweeted
Every single person you meet in your daily life is fighting a hidden battle that you know nothing about Send them a Zoom meeting link for an 8 AM call on Monday morning to finish them off
143
2,375
105
35,549
896,420
Casperinous retweeted
At #hexacon - Calif.io demos iOS 27.0.1 untethered jailbreak on iPhone 18 Pro
94
328
81
2,208
554,910
Casperinous retweeted
Having previously posted that this was an interesting rabbit hole, a spam account liking that post a couple of weeks ago reminded me that I never actually explained why. So here's a brief version - maybe something we should blog about in more detail in the near future...
2
15
1
45
4,110
Casperinous retweeted
Wrote up my attempt at deobfuscating Luraph-protected Lua: lifting Lua to LLVM IR , then using AI + partial evaluation to strip the VM, and checking it against the original. nac-l.github.io/2026/10/12/d…
2
31
2
151
7,303
🦠 ThreatLabz has identified an x64 variant of GuLoader. The sample uses similar anti-analysis methods as GuLoader x86 samples. However, the following anti-analysis features have been added: - Huge stack frames in order to break IDA's decompilation. - Direct syscalls to call Windows native API functions. GuLoader chooses a different syscall address that is chosen randomly. - The registered Vectored Exception Handler (used to obfuscate the control flow) points to a location within NTDLL's memory region, making the exception-driven control flow appear legitimate. - Use of the Windows API function TrySubmitThreadpoolCallback to execute callbacks and avoid detection. Sample hash: dc5021a0e1d00fb1da9eef2b5d5ac3f9a5906e5c3c17470807c3845083ac5a6d Malicious Google Drive URL: hxxps://drive.google[.]com/uc?export=download&id=1Zynj-CzTxaYpQr5aniRGE5gUqs26k-t5
14
1
43
2,856
Casperinous retweeted
I can’t tolerate people being cruel to everyone else dealing with their emotions and the implications of ai changing everything about their work (and life!) have some decency man
22
29
1
517
15,747
Casperinous retweeted
This semester I am teaching a new course on Agentic Systems, which explores the full stack of AI systems: from building and optimizing agent harnesses to serving and optimizing LLM serving infrastructure. Students learn by building, measuring, and optimizing real agents and systems, balancing capability, latency, and cost. Our goal is not just to teach today's AI technologies, but to prepare students to build tomorrow's. All lectures, readings, and assignments are publicly available at cs2680.seas.harvard.edu
67
429
13
3,820
145,680
Casperinous retweeted
The final post of my decompilation project is out. 83% of the functions are byte exact. We're hitting diminishing returns. The game runs flawlessly, so we're calling it done here. For obvious reasons, no code will be shared. momo5502.com/posts/2026-10-0…
39
51
14
553
86,850
Casperinous retweeted
Check out our new VMProtect (v3.0 - v3.10 latest) unpacker 👀 More to come…
7
17
1
170
16,989
poteto's cursor profile is public. She's used 1.5T tokens in the last 30 days. Assuming an avg rate of $8/M that's $384k/day or $11m. In the last 30 days. Per year thats $132M. Let's assume Cursor has some insane deal with Anthropic and gets 90% off. That comes out to $38k/day or $1.1M in Sept/Oct. $13m will get me 26 sassy staff-level all stars for $500k/yr
Readers added context they thought people might want to know
The Cursor profile at cursor.com/@lauren shows 1.5T tokens used. Cost estimates at $8/M do not account for prompt caching common in agent coding, where cache reads are billed at ~10% of input rates per analyses of Anthropic pricing. This results in lower effective costs. cursor.com/@lauren chudi.dev/blog/claude-pr… carlosaragon.online/blog/anthropic…
274
91
93
3,133
471,388