Robert Musser
@r_o_b_e_r_t_1
I like cool stuff.
You might like
OPEN-SOURCES ALL MODEL CHECKPOINTS AND TRAINING LOGS REFERENCED IN THE PAPER HF: huggingface.co/collections/ji… PAPER: arxiv.org/abs/2511.03276 MEGADLMS: github.com/JinjieNi/MegaD…
Diffusion Language Models are Super Data Learners… now on arXiv with MegaDLMs, the full large-scale training framework (6.1K H100s, 462B-param run, 47 % MFU). Supports diffusion and autoregressive LMs, dense and MoE architectures, FP8/BF16/FP16 precision, and multi-axis…
vLLM Sleep Mode 😴→ ⚡Zero-reload model switching for multi-model serving. Benchmarks: 18–200× faster switches and 61–88% faster first inference vs cold starts. Explanation Blog by @EmbeddedLLM 👇 Why it’s fast: we keep the process alive, preserving the allocator, CUDA graphs,…
Also out: R-HORIZON. It composes interdependent chains across math, code, and agent tasks to test real long-horizon reasoning. Top models degrade rapidly as horizon grows: DeepSeek-R1 falls from 87.3% to 24.6% at 5 linked problems, R1-Qwen-7B drops from 93.6% to 0% at 16.…
LongCat isn’t a one-off. It’s part of a deep, coordinated research push that’s been unfolding quietly inside Meituan, the same research platform behind Flash-Thinking for large-scale reasoning, CodePlot-CoT for visual math, M4V for multimodal diffusion, BNPO for stable…
a periodic reminder: there 👏 is 👏 no 👏 such 👏 thing 👏 as 👏 private, end-to-end encryption 👏 with 👏 meta-ai-in-the-middle 👏
Demystifying Synthetic Data paper: arxiv.org/abs/2510.01631 Data Mixing & Phase Transitions paper: arxiv.org/abs/2505.18091
Meta just ran one of the largest synthetic-data studies (over 1000 LLMs, more than 100k GPU hours). Result: mixing synthetic and natural data only helps once you cross the right scale and ratio (~30%). Small models learn nothing; larger ones suddenly gain a sharp threshold…
I've been researching the Microsoft cloud for almost 7 years now. A few months ago that research resulted in the most impactful vulnerability I will probably ever find: a token validation flaw allowing me to get Global Admin in any Entra ID tenant. Blog: dirkjanm.io/obtaining-glob…
I just hope they infected "left-pad", "is-number", "is-odd", "is-even" packages
It’s pretty fucking straightforward here! thebulwark.com/p/stock-tradin…
It’s genuinely insane that it hasn’t been shut down yet. Like, just open flouting of a law passed by Congress, signed by a president, and upheld 9-0 by the Supreme Court.
> be Google in 2017 > small team drops “Attention Is All You Need” on arXiv > execs nod politely, go back to selling ads for socks > let Transformer gather dust for 5 yrs like a vintage Beanie Baby > be Noam Shazeer, OG wizard > quits, builds AI-boyfriend app…
All #OrangeCon2025 talks are now online! Watch them on our YouTube channel: youtube.com/@OrangeCon
> be vibe coder > 2025: “I'm gonna vibe-code the next unicorn, it's a billion-dollar vibe, bro” > grab xxx.ai, .ai costs more than my rent > subscribe to every tool in existence, $1000 bucks gone > tweet demo GIF, caption “built in 3 hours, no cap”, await…
There's a sick linenoise article by @iximeow in @phrack 71 called "Learning An ISA By Force Of Will", where ixi goes from unknown binary blob, to manual instruction decoding, to figuring out control flow, and gives a critique of the RE'd ISA. phrack.org/issues/71/3#ar…
How do you program an unknown CPU? The original specs are gone; no compilers exist, and the ISA is completely unrecognized. It happens more often than you think, behind very closed doors. It's almost always military hardware.
The Great Firewall of China (GFW) today experienced the largest internal document leak in its history. More than 500GB of source code, work logs, and internal communications have been exposed, revealing details about the development and operation of the GFW. The leak originated…
Exciting times. I'm publishing Dittobytes today after presenting it at @OrangeCon_nl ! Dittobytes is a true metamorphic cross-compiler aimed at evasion. Use Dittobytes to compile your malware. Each compilation produces unique, functional shellcode. github.com/tijme/dittobyt…
REPO: github.com/allenai/OLMoASR RELEASE: allenai.org/blog/olmoasr MODEL: huggingface.co/allenai/OLMoASR SET: huggingface.co/datasets/allen…
United States Trends
- 1. $PUFF N/A
- 2. #FanCashDropPromotion N/A
- 3. Good Friday 47.4K posts
- 4. #FridayVibes 3,588 posts
- 5. $apdn $0.20 applied dna N/A
- 6. $SENS $0.70 Senseonics CGM N/A
- 7. $LMT $450.50 Lockheed F-35 N/A
- 8. Publix 1,331 posts
- 9. Happy Friyay 1,000 posts
- 10. #FridayFeeling 2,246 posts
- 11. #fridaymorning 1,408 posts
- 12. #PitchYaGame 1,073 posts
- 13. RED Friday 3,036 posts
- 14. Finally Friday 3,787 posts
- 15. Elise Stefanik 3,205 posts
- 16. John Wayne 1,401 posts
- 17. yeonjun 307K posts
- 18. Sydney Sweeney 106K posts
- 19. Kehlani 14.7K posts
- 20. Out The Window 11.3K posts
Something went wrong.
Something went wrong.