Bill Demirkapi
@BillDemirkapi
Followers
22K
Following
3K
Media
195
Statuses
1K
Security Researcher. Passionate about OS internals, Cloud Security, and everything in between. Opinions are my own.
Boston, MA
Joined July 2017
Thanks to everyone who attended my talk! @offensive_con was an incredible experience. Couldn't make it? Don't worry- here are my slides! Recommend them to anyone wanting to learn more about Secure Boot's third-party risks.
Ecstatic to share that my talk was accepted to Offensive Con. Excited to have an excuse to visit the city I was born in. Hope to see you there!
3
15
71
real red pill is automating the decomposition of all binary formats, not just executables. think last year of PDF data mining but now scale it to every file format ever.
0
0
0
automating reverse engineering would increase training data by an order of magnitude. consider decomposing every closed source application in the world. you could start with easily reversible formats like JARs/APKs before going after stripped executables.
new blog post. "There Are No New Ideas In AI. Only New Datasets". in which i summarize LLMs in exactly four breakthroughs and explain why it was really *data* all along that mattered. not algorithms
2
2
13
RT @ylecun: AI doomer: "OMG, I told my AI assistant that I'll shut it down and it told me to kill myself 😱😱😱".AI assistant: .
0
318
0
RT @ludwigABAP: I suggest we find the people responsible for post-training LLMs into emoji-slop idiots and we simply use metal bars on thei….
0
73
0
trying to figure out how the hell im gonna get two 240v power connections to this mf living in a rental so I asked chatgpt. "cook dinner on induction for now; swap cords when you need the server" is crazy ngl ðŸ˜ðŸ¥€
3
0
10
Stanford's @HazyResearch group recently published a private AI inferencing implementation with bold claims about encrypting user data end-to-end, from client to GPU enclave. Nearly all their claims are provably false. I do not say this lightly. Read more:
3 months ago, Stanford's Hazy Research lab introduced Minions, a project that connects Ollama to frontier cloud models to reduce cloud costs by 5-30x while achieving 98% of frontier model accuracy. Secure Minion turns an H100 into a secure enclave, where all memory and
1
3
23
RT @thdxr: the problem with all of these predictions is none of these people seem to have seen what the inside of enterprise organizations….
0
45
0
Windows Recall was one of my coolest security architecture projects! Super proud of the team for implementing the non-trivial enclave separation and anti-abuse controls. I remain skeptical of VSM's defensibility, but Recall is in a far better state than it was a year ago.
David "DWIZZZLE" Weston (@dwizzzleMSFT), CVP, Security at Microsoft, kicked off Day 1 of BlueHat India with a keynote on the security implications of on-device AI and the evolving threat landscape. He outlined three critical categories of AI risk:.1. Data attacks: The biggest
0
0
7
openai really cooked with o1-pro. yea, I need to remortgage my home now, probably get a 401k loan at this rate, but damn does it work. if hallucinations are an issue with the o3/claude 3.7 max/gemini 2.5 pro max models, highly recommend.
0
0
7
as I've trusted my gut more with difficult choices, slowly starting to realize overthinking may bring greater short-term confidence, but intuition wins in the long run.
1
0
5
RT @WillManidis: the yankees hired a single smart guy to think about baseball bats for a year and he fundamentally changed the game forever….
0
1K
0
RT @taviso: You can now jailbreak your AMD CPU! 🔥We've just released a full microcode toolchain, with source code and tutorials. https://t….
0
523
0
RT @h0mbre_: don't half-ass/dabble in things because you're worried that if you actually did try you would fail and you need to protect you….
0
18
0