never doom. @FactoryAI ambassador 24 year old nerd
hyderabad
Joined February 2022
- Tweets12.2K
- Following940
- Followers4.9K
- Likes28K
lmao i made more progress with $2.5 and abliterated deepseek (thanks @skeptrune and @morphllm for the access) in 2 hours than i did with 2 days of frontier models because of their shitty cyber guardrails
goodnight
the fuck do you mean you WON'T train the model properly and entirely just drop it? why? seems very wasteful
if a model shows any misaligned behaviour you just kill the run and go back to a previous checkpoint?
hope they know what they're doing
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
DeepSeek v4.1 Flash 1017 is open-weight cyber SoTA! Since Kimi K3.1 Pro is nerfed for cybersecurity, I don't think we'll get a good public release like this until.. GLM 5.5?
DeepSeek V4.1 Flash 1017 is the new #1 openly-available model in the Artificial Analysis Cyber Index. Other proprietary models may offer better cyber defense performance but are restricted access
@deepseek_ai recently released DeepSeek V4.1 Flash 1017, an update to their flagship open-weights model DeepSeek V4.1 Flash. With an Artificial Analysis Cyber Index score of 57%, DeepSeek V4.1 Flash 1017 is the #1 openly-available model for cyber defense capabilities.
The Artificial Analysis Cyber Index measures capability on agentic cyber defense tasks and is composed of three benchmarks: CWE-Bench-AA, DeepsecBench-AA and CyberGym-E2E-AA. The Cyber Index primarily evaluates models through their publicly accessible APIs, reflecting the cyber capabilities available to developers.
Many models, particularly proprietary models from OpenAI, Anthropic and Google, may have stronger defensive cyber capabilities but encounter safety refusals that reduce their overall scores. GPT-6 Sol (Daybreak Blue, max), available through OpenAI’s trusted-access program, leads the Index, scoring 32 points above its public counterpart driven by fewer refusals. DeepSeek V4.1 Flash 1017 encounters no safety refusals.
Key results:
➤ DeepSeek V4.1 Flash 1017 has the highest score among publicly-available models on the Artificial Analysis Cyber Index and #2 position overall. DeepSeek V4.1 Flash 1017 (max) scores 57% ahead of the previously leading publicly-available models Grok 4.7 (xhigh) and MiMo-V2.6-Pro by 1 point. It places #2 overall, with the trusted-access model GPT-6 Sol (Daybreak Blue, max) leading by 12 points with a score of 69%
➤ DeepSeek V4.1 Flash 1017 improves on its predecessor DeepSeek V4.1 Flash by 16 points. DeepSeek V4.1 Flash 1017 (max)’s 16-point improvement from 41% to 57% is driven by a substantial 53-point improvement on CyberGym-E2E-AA from 23% to 76%. On CWE-Bench-AA and DeepsecBench-AA it scores 69% and 25%, compared to 68% and 31% for DeepSeek V4.1 Flash (max)
➤ With a Cost per Task of $0.42, DeepSeek V4.1 Flash 1017 sits on the Capability vs. Cost Pareto frontier. DeepSeek V4.1 Flash 1017 costs $0.42 per task, more than 2x MiMo-V2.6-Pro (56%) at $0.18 per task but 4x cheaper than GPT-6 Sol (Daybreak Blue, max, 69%) at $1.77 per task
oh my god bruh
Now live: abliterated-model-large-flash (our version of DeepSeek-V4.1-Flash-1017). It is a cheaper alternative to our GLM-5.3-based model for product owners and defenders who need access to the most cyber-capable models.
- 1M context
- Vision
- Zero input/output retention or training
We believe paying customers should be in control, and over the past few months, leading experts and major brands have adopted our models.
abliteration.ai
someone with an idle/not very used decked out mac studio give me ssh access and let me setup some variant of DS v4.1 Flash abliterated/GLM 5.3 Flash abliterated
you will need it in these trying times (yes please)
UGH my claude is hitting cyber guardrails reading the handoff IT WROTE yesterday while looking for security vulns
im having to use sol to 'soften' and safe guardrail claude's own handoff to be able to continue it's work
it's impossible to work with in subagents because all of them will hit guardrails separately and claude will either completely fail to steer it back into not hitting guardrails or it will hit the guardrail itself looking at what the subagents are doing, so i'm stuck with one agent, having to prevent it from crashing into guardrails, having it apply patches, confirming with sol that it is indeed safe, merge, push, test staging/prod, repeat for the last 24 hours
ive run out of usage in my droid sub and i'm at the verge of just upgrading there and spamming glm 5.3 but i'm almost done for the weekend and after all the critical shit is done ill just have glm running over the weekend to figure out if there's some weird way it can chain a whole set of weird decisions into another exploit
im losing my mind with all these refusals
i do not want to pay big boy money for @abliteration_ai
other glm 5.3 (flash) abliterated providers are weird
i can only run 120b param models and im not finding any new vulnerabilities and am barely confident in my patches
what do?
i will assume this is a lot of Fable 5.5 at work and the engineering prowess i'm going to have access to SOON™️ has me excited
all that just to be unable to make any security patches is funny but FUCK man i need this model
cheaty retweeted
You asked for Xbox 360, so here it is.
Naruto: Rise of a Ninja now runs natively on PS5, 4K output and the same 30/60 FPS as the original.
im losing my mind with all these refusals
i do not want to pay big boy money for @abliteration_ai
other glm 5.3 (flash) abliterated providers are weird
i can only run 120b param models and im not finding any new vulnerabilities and am barely confident in my patches
what do?
people who didn't use their subscription all the way but kept hitting 5 hour limits is who this is for obviously
but im not a fan of it, bad a/b test to do
hard problem to solve without entirely changing how limits work, i still think we 3-4 hours limit not 5 hours
