Hask (@Hask_Sec) profile picture

eW91dHUuYmUvUnNIWXoxWXBRS0k=

4e 6f 77 68 65 72 65
Joined January 2012
Photo by Hask (@Hask_Sec)Photo by Hask (@Hask_Sec)Photo by Hask (@Hask_Sec)Photo by Hask (@Hask_Sec)Photo by Hask (@Hask_Sec)Photo by Hask (@Hask_Sec)Photo by Hask (@Hask_Sec)
Phrack 73 just dropped. Tell your SOC you'll be busy phrack.org/issues/73/1
25
595
48
1,822
103,046
Introducing Qwen3.8-27B-pi-adapted 🤗 We adapted Qwen's model specifically for the pi coding agent using reinforcement learning and a LoRA adapter. The results, measured as pass@1: 🚀 SWE-bench Pro (Hard 51): 29.4% → 62.7% 💻 Terminal-Bench 2.1: 39.5% → 55.8% ⚡ 55–62% fewer tool calls per task This is the power of open models: you can adapt them to YOUR tools and YOUR use cases. Let’s build! 💛 Open weights: huggingface.co/harness-hub/Q… Methodology: huggingface.co/blog/harness-…
268
565
148
5,255
334,426
Hask retweeted
Polestar 0.4.0 is out. The Riverglades and Shen'dralas are now part of the 1 to 60 route for both factions, both routes were walked again zone by zone, and the settings window no longer stutters. Free and open source, for WoW Forever.
66
22
4
883
78,678
Hask retweeted
I trained a text-to-speech model that has never heard a real human voice. 4.5M parameters. 18 MB. Runs on a CPU. Meet tiny-voca
293
547
126
8,407
531,724
Great talk by @PeteMarkowsky at Reddit SnooSec: using local models to block bad actions in realtime. You can derive intent from local session files, and he compared Apple’s afm 3 with Qwen. The results are outrageous. 2 seconds for afm vs Qwen’s 200ms.
12
8
1
91
5,419
Tool: transfer data through sound, including INAUDIBLE frequencies! Use it here: chirp.hex.dance Works offline, data not sent to any server: it’s an in-browser modem supporting frequencies between 18k and 19k Hz, which most humans can’t hear (but your cat can!)
165
394
69
3,040
113,792
Hask retweeted
being really good with computers is an absolutely incredible skill because it lets you make your own life unnecessarily complicated in ways normal people could never imagine
79
1,533
69
19,801
249,100
Update: I was correct. Pwn AI nerds don't have elite ball knowledge and don't know anything about goop As is tradition, it was something stupid. Stay in your lane, dorks. This is goop city
No, Binance was not most likely compromised from a web browser zero day chain. As is tradition with large compromises, it is almost certainly something ridiculously stupid, like the DevOps guy trying to get Fortnite cheats, or they called someone with a shitty American accent and said they need to log into their machine for something
52
151
9
2,621
57,282
BREAKING: Revolut moved ~2,000 engineers onto self-hosted GLM for coding and pushed Claude Code and Codex to the exception list. Per David Abitbol, who leads AI ops there: 56 Nvidia B200s, a blended 30 cents per million tokens, roughly a tenth of what they paid Anthropic and OpenAI. Engineering lead time improved 70%. Here's the part that matters: the gains showed up when tokens stopped being rationed. Engineers quit saving AI for the hard tickets and pointed it at everything. Frontier models become the exception, open weights become the default. That's a pricing problem for every lab heading to IPO. h/t @OliverMolander Would you move your team off frontier APIs if the bill dropped 90%?
306
696
385
7,813
1,017,825
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math. It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training. We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵 Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math. rikiyat.github.io/blog/weight-sp… github.com/RikiyaT/lm_und… huggingface.co/Qwen/Qwen3.5-0… huggingface.co/RikiyaT/Qwen3.…
2,366
7,256
5,753
55,941
10,068,103
No C2 allowed? Beacon over Microsoft Teams. Reverse engineering Teams' meeting APIs and WebRTC data channels to run a C2 over Microsoft traffic that the infrastructure already allows. A post by Clément Labro (@itm4n) Source: blog.scrt.ch/2026/04/21/from… #redteam #blueteam
1
80
1
344
11,750
By the way, in case anyone was still wondering whether they are completely useless, Binance holds SOC 2 Type 2, ISO 27001, and several other certifications.
161
456
90
4,796
254,727
What it feels like when you start using a Chinese model and it actually just does what you want without a fucking condescending moral lecture.
273
1,894
109
35,991
598,831
“Alignment for security? Is this a joke? Are you a child?” Truly amazing work by @sherrod_im
23
41
6
252
12,610
Last night an AI agent performed brain surgery on a mouse. We gave it 19 gene therapy vectors in a spreadsheet, a mouse model of epilepsy, access to a motorize surgery rig with cameras and one goal: reduce seizures. It picked the drug. It picked the brain region. It derived the coordinates from three papers. Then it drove the stereotaxic arms into the brain and injected 250 nanolitres while watching the entry site through the rig camera for bleeding. A licensed vet gated every irreversible step and held a physical kill switch to stop the AI agent. Thread on what it decided, what it saw, and the moment it refused to let me override it. [Writing this on the spot out of excitement, but I've been 18hrs in front of my laptop coding and coordinating this, so keeping it brief and descriptive, more cohesive thoughts on this soon...]
311
1,293
700
9,700
1,687,694
Hask retweeted
found some neurons of your own you want to share with the world? maybe a particularly funny persona or a super specific fixation? share them here! neuronary.org/
if you have a qwen3.5 4b around, please try steering neurons L14 N8949 +1 and L13 N7956 -1 now you have evil qwen recipe: dose × median(‖resid‖ leaving layer L, tokens 1:) / ‖W_down[:, n]‖` at the down_proj at every input
25
80
18
793
45,311
Introducing the "Grim Prompter" one of the largest collection of open source threat modeling prompts on the planet. 😎 I built Grim Prompter to help developers threat model with Claude Code, Codex, or any coding agent. 36 architecture workflows and 29 method and analysis prompts! MIT licensed. Use it, improve it, share it. Security tools should help everyone. github.com/jmanico/grim-prom…
31
40
4
198
13,781
if you have a qwen3.8 27b around, try steering L34 N3838 +0.75, now you have evil qwen 27b for extra evil add L38 N17180 +0.75 same recipe as @_lyraaaa_: dose × median(‖resid‖ leaving layer L, tokens 1:) / ‖W_down[:, n]‖ at the down_proj input, every token
138
124
127
2,017
603,427