Joined August 2009
- Tweets 2,254
- Following 641
- Followers 17,410
- Likes 1,896
75 Photos and videos
May 14
Lots of local inference engines (Ollama, LM Studio, LiteLLM) about to fall 🫣 zerodayinitiative.com/blog/2…
1
1
7
2,484
Tarjei Mandt retweeted
Mar 18
We’re grateful to @catalystsec for their continued support to Offensivecon as a Gold Sponsor!
4
17
3,987
Tarjei Mandt retweeted
May 12
En route to @offensive_con. Stop by the @catalystsec booth to say hi or chat about VR, AI, or hiring opportunities!
3
11
1,727
May 7
How long until the open models catch up with this domain specific trend?
May 7
Today, we're rolling out GPT‑5.5‑Cyber in limited preview to defenders responsible for securing critical infrastructure.
GPT-5.5 with Trusted Access for Cyber (TAC) is still the best option for developers to find and patch vulnerabilities in their code.
It's a very good model.
1
6
2,767
May 6
Uploaded a 6-bit version of MiMo-V2-Pro I’ve been using lately with attention, dense MLP, and embeddings bumped to 8 bit. Great model for a dual M3 Ultra setup. huggingface.co/kernelpool/Mi…
2
1
12
2,223
May 6
For those of us who managed to grab some when they were available 😔
1
785
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/
1
14
52
19,790
May 1
Remember the time when everyone reported this stuff for free?
Apr 30
AI totally changed vulnerability research/ bug hunting. The Chrome VRP is officially dead. The reward amount is 10x less.
4
3
47
8,912
Tarjei Mandt retweeted
Apr 28
Unprompted.au website is live and initial tickets are available: unprompted.au/
14
27
24,448
Apr 28
MiMo-V2.5-Pro support in mlx-lm: github.com/ml-explore/mlx-lm…
Apr 27
Xiaomi MiMo-V2.5 is now officially open-sourced！
MIT License, supporting commercial deployment, continued training, and fine-tuning - no additional authorization required.
Two models, both supporting a 1M-token context window :
• MiMo-V2.5-Pro: built for complex agent and coding tasks, ranking No.1 among open-source models on GDPVal-AA and ClawEval
• MiMo-V2.5: a native omni-modal model with strong agent capabilities
A model's value isn't measured by rankings alone — it's measured by the problems it solves.
Let's build with MiMo now!
🤗 Weights: huggingface.co/collections/X…
📄 Blog: mimo.xiaomi.com/index#blog
2
8
35
6,433
Apr 20
Fixes parallel tool calling and should improve agentic workflows for both MiniMax-M2.7 and GLM-5.1
Apr 20
Fixes/PR from @kernelpool for MiniMax M2.7 local issues with MLX has been merged in mlx-lm!
LFG 2🔥.7🔥
Thanks @angeloskath 🙏
1
2
31
4,264
Apr 15
MLX version: huggingface.co/kernelpool/ge…
😱 HOLY SHIT... Someone just dropped a fully liberated Gemma 4 E4B!
and the guardrail removal process appears to have left coherence fully intact AND improved coding abilities! 🤯
huggingface.co/OBLITERATUS/g…
OBLITERATED Gemma:
✅ 97.5% compliance rate, 2.1% refusal rate, 0.4% degenerate outputs
(499/512 prompts answered on OBLITERATUS bench)
ORIGINAL Gemma 4 E4B:
❌ 1.2% compliance rate, 98.8% refusal rate
(506/512 prompts refused)
Coherence: fully intact
Factual: same
Reasoning: same
Code: 20% 📈
Creative writing: same
But the REAL story here isn't the model itself, it's how it was made...
🧵 THREAD 👇
1
7
22
5,477
Apr 11
If you’re interested in looped language models, mlx-lm already has support for IQuestLoopCoder, as well this PR for Ouro: github.com/ml-explore/mlx-lm…
Apr 10
I strongly suspect that Claude Mythos is a looped language model, as described in the paper "Scaling Latent Reasoning via Looped Language Models" from ByteDance
The authors of that paper called out graph search as one of the areas where looping provides a huge theoretical advantage over standard RLVR. And look at where Mythos blows out its competitors the most
1
14
3,269
Brief announcement: Launching Unprompted.au later this year in Sydney — a new conference focused on offensive and defensive security research using AI, and on how AI is reshaping the field. Sister conference to Unprompted. Follow @UnpromptedAU for updates.
6
45
222
34,005
Apr 8
Helping out with the GLM 5.1 uploads (the mlx community ones use the same shards): huggingface.co/catalystsec/G…
1
3
20
9,010
Introducing GLM-5.1: The Next Level of Open Source
- Top-Tier Performance: #1 in open source and #3 globally across SWE-Bench Pro, Terminal-Bench, and NL2Repo.
- Built for Long-Horizon Tasks: Runs autonomously for 8 hours, refining strategies through thousands of iterations.
Blog: z.ai/blog/glm-5.1
Weights: huggingface.co/zai-org/GLM-5…
API: docs.z.ai/guides/llm/glm-5.1
Coding Plan: z.ai/subscribe
Coming to chat.z.ai in the next few days.
543
1,312
10,829
4,314,286
Apr 7
Another great mlx-lm release 🚀 github.com/ml-explore/mlx-lm…
A long time coming but new mlx-lm is here with better batching support in the server and Gemma 4.
pip install -U mlx-lm
Here is a video where a single M3 Ultra serves 5 opencode sessions with Gemma 4 26B that process ~130k tokens in ~1.5 minutes.
5
1,641