Is this the first time Meta and Apple work together on CEO level? MLX is the preferred local inference platform for Meta, and we get 0-day support for Muse Spark Open in mlx-lm with vision support.
A lot of people use third party inference engines other than mlx-lm, and they will all benefit from the optimization works for Muse Spark Open in MLX. Which is an important meaning of the mlx-lm project: to provide a good foundation for supporting new models, not to compete with open source community.
For the technical details, our new blog is a great read: Local and Distributed Large LLM Inference with MLX and RDMA on Mac Studio machinelearning.apple.com/re…
Replying to @AIatMeta
To help make advanced models more accessible, Muse Spark Open is optimized to run on two Apple M5 Ultra Mac Studios with the latest MLX framework.
It also runs on leading inference engines like SGLang, vLLM, TensorRT-LLM, Hugging Face Transformers, llama.cpp, and Unsloth with optimizations for AMD, Apple, Intel, and NVIDIA silicon.
Developers can further customize the model with PyTorch, including TorchTitan for fine-tuning.
Tarjei Mandt retweeted
Imagine the amount of Java crashes this poor LLM must have in its training set that it decided 'NullPointerException' deserves its own token
Tarjei Mandt retweeted
I put my @UnpromptedAU slides up at justdionysus.github.io/slide… — a bit of reflection on exploit development in the age of AI. My TL;DR is keep pushing to understand complex things, be honest with your own understanding, and use AI as a power tool to increase pace and depth.
Tarjei Mandt retweeted
Can we please stop posting about rogue AI and how it’ll kill us and go back to posting about all the cool vulns it is finding?
Tarjei Mandt retweeted
Qwen 3.8 Flash Next support in DwarfStar is ready!
@kernelpool + @ddalcu + @Spangler3000 + me combined. Prefill with this architecture is incredible! Flat up to 262K!
PR here: github.com/antirez/ds4/pull/…
Trying to push decode with MTP even more, and working on Q2 Model for 64GB machines as foreseen by @antirez 🚀
Tarjei Mandt retweeted
I only see "ExploitBench: 100%"
Chubby♨️@kimmonismusSep 3Germany
GermanyConnected via Germany App StoreAccount-level information, not a live location or per-post device.
Full benchmarks. Absolutely insane.
ARGI-AGI 3, from 7.8% to 98.6%
Exciting announcement: I am launching a new AI-focused company today named Umbriel (@Umbriel_AI)! Excited to do some cool research in the space 😎
Introducing Umbriel: a new company focused on LLM research and engineering, particularly the development of advanced AI-driven cybersecurity pipelines, as well as broader research into LLM tech.
We have some exciting things planned. Watch this space!
umbriel.ai/
Tarjei Mandt retweeted
Ok! The schedule is live as! Check it out: unprompted.au
Introducing GLM-5.3: Built to Code. Ready for Cyber Defense.
- Top-tier coding and agentic capabilities, achieved through post-training on the 743B base model
- A major leap in cybersecurity, setting a new standard among open models
Tech Blog: z.ai/blog/glm-5.3
Tarjei Mandt retweeted
I understand that people are optimistic about all of software engineering being automated after the jump of capabilities in the last 10 months. But ... I don't know. I don't see production-grade code happening without a lot of steering just yet, even for simple things.
Tarjei Mandt retweeted
Today we're also opening the weights for Muse Glimmer, a great 30B parameter dense model that can run locally. Soon we'll also release the weights for Muse Spark 1.2, our latest foundation model. Meta is a strong supporter of open source and I'm proud of these releases. Congrats to @alexandr_wang and the MSL team for all your great work on these models.
Tarjei Mandt retweeted
I just asked Opus 5 to describe me in 2 sentences for a conference. That triggered the cybersecurity safeguard and the model was downgraded to 4.8
Tarjei Mandt retweeted
if you're doing genuinely interesting work where AI meets security research, unprompted.au wants your talk. Sydney, Sep 18–19. i'm on the program committee and the CFP closes July 31 — send it 👉 sessionize.com/au-unprompted @UnpromptedAU
Tarjei Mandt retweeted
We’re grateful to @catalystsec for their continued support to Offensivecon as a Gold Sponsor!
Tarjei Mandt retweeted
En route to @offensive_con. Stop by the @catalystsec booth to say hi or chat about VR, AI, or hiring opportunities!
CFP is open!! Submit your cool talks about AI hacking/defense/core tech. Gonna be awesome!! ☺️
The Unprompted.au CFP is officially OPEN! If you are doing cool stuff with AI in offense, defense, or working on core AI tech (from frontier models to open source LLMs), we'd love to hear from you! Submit here: unprompted.au/




























