Luciano Remes

31 Followers
11 Following
24 Posts
SWE @ Palantir | systems researcher
site:https://www.lremes.com/
twitter:https://twitter.com/cybergenik

I keep hearing people saying that R1 is Open Source: "All the model code is open".

This is an absurd mischaracterization of what's happening. All the code is NOT open source, most importantly the training code is NOT open source.

What actually happened:
- Deepseek released a paper outlining how they built the model: https://arxiv.org/pdf/2501.12948. A lot of the claims made in it are dubious at best, pre-training and training steps are extremely opaque, and I haven't seen an independent researcher that's been able to reproduce the cost of training claims. After reading the paper, I don't doubt there was probably some degree of savings in the training process. But I wouldn't trust the figures blindly, China has a massive incentive to lie about these kinds of things.
- Deepseek published this repo, which basically just contains the paper: https://github.com/deepseek-ai/DeepSeek-R1
- Deepseek published model WEIGHTS in a hugging face repo: https://huggingface.co/deepseek-ai/DeepSeek-R1

I think it's very irresponsible to continue to spread this wide misconception that models like llama and Deepseek-R1 are Open Source. They're not, you won't be able to train it because you don't have the training code and a complete dataset, full stop.

What can you do with the weights? You can load them into a framework like PyTorch and run inference. But you have no idea how the weights were generated, which is maybe the most important part.

My first year at #defcon32 was absolutely wild.

Grateful to everyone who came to our demo lab about Hopper. Had some great questions!

Met some cool people, had some hilarious, very interesting, and slightly unhinged conversations. Lost a bit of money gambling, had one too many drinks, went to the Illuminati party (the DJ was next level). Got exposed to some new ideas, met some old friends, made new friends.

To top it all off, the pool party Saturday night was perfect, beach ball fight went hard. I think our side won, ggs.

If you’re min-maxing for unique interactions, it’s GTO to come back next year. Until next time @defcon !

I’ve been selected to present at DEFCON 32! I’ll be doing a Demo Lab on my thesis project Hopper: Distributed Fuzzer with my co-presenter Wade Cappa

We’ll be going over key ideas, results, and demos on using it to run distributed fuzz campaigns. Hope to see you all there!

Huge shoutout to my advisor @ricci for his guidance during my time at FLUX.

https://github.com/Cybergenik/hopper

#DC32 #DEFCON #fuzzer #security

GitHub - Cybergenik/hopper: Coverage-Guided Greybox Distributed Fuzzer

Coverage-Guided Greybox Distributed Fuzzer. Contribute to Cybergenik/hopper development by creating an account on GitHub.

GitHub

Been thinking a lot about what it means to be a creator. So I wrote a small poem:

https://www.lremes.com/posts/creator/

Luciano Remes | The Creator

What is a creator? A founder, an architect, a painter? Someone who creates? A creator is vigilant of flaws in the world, they drive him mad…

I love game theory, so I wrote a blog a while ago about thinking about stuff through the lense of game theory:
https://www.lremes.com/posts/gto/
Luciano Remes | Don't be Outcome Oriented

Life is a multiplayer imperfect game, there are other players in this game, each focused on optimizing their own set of objectives. A…

Whenever explaining my research, I always inevitably get asked "What is a Fuzzer", so I wrote a blog about it:

What is a Fuzzer?
https://www.lremes.com/posts/fuzzing/

#fuzzers #fuzzing #security #AFL #systems

Luciano Remes | What is a Fuzzer?

Inspiration: I wrote my undergrad Thesis on a new distributed fuzzer called Hopper that I built, and I always get the question: "What is a…

The new draft for the #rustlang trademark copyright is… a bit wack.

Thankfully it’s just the first draft, and they’re taking feedback. I hope this changes substantially, some of the currently proposed policies feel a bit overreaching.

https://docs.google.com/document/u/0/d/1ErZlwz9bbSI43dNo-rgQdkovm2h5ycuW220mWSOAuok/mobilebasic

Call for Comment: Rust Trademark Overview, FAQ and Policy April 2023

Fuzzing readelf on a 48 core machine with my #distributed fuzzer Hopper:

https://github.com/Cybergenik/hopper

I have 4 more of these doing the same thing, lets see if I find something interesting.
#cloudlab #fuzzers #gnu

GitHub - Cybergenik/hopper: Coverage-Guided Greybox Distributed Fuzzer

Coverage-Guided Greybox Distributed Fuzzer. Contribute to Cybergenik/hopper development by creating an account on GitHub.

GitHub
One of the most coveted types of vulnerabilities is a Linux kernel 0-day with arbitrary code execution. But after decades, how many kernel 0-days do you think are still out there?
#zeroday #kernel #linux
0-10
22.2%
50-100
22.2%
100-500
37%
1000+
18.5%
Poll ended at .

If I'm not at a terminal (seldom), I'll sometimes use linux.die.net to look at man pages, overall a great website.

Recently I was looking through the watchdog man pages. I noticed these weird ASCII glyphs. I'm guessing these are supposed to be bullet-points, maybe some ASCII parsing that's getting messed up. Does anyone know what's going on?

#linux #man-pages
https://linux.die.net/man/8/watchdog

watchdog(8): software watchdog daemon - Linux man page

watchdog is such a daemon. It opens /dev/watchdog, and keeps writing to it often enough to keep the kernel from resetting, at least once per minute. Each ...