Dr. Wesley McGrew (@McGrewSecurity) header
Dr. Wesley McGrew (@McGrewSecurity) profile picture

https://teapawt.net/t.co/J2fRTxUccG | Senior Cyber Fellow at @martinfederal | Offensive Computer Security; AI; Photography; DJ

Joined February 2008
Photo by Dr. Wesley McGrew (@McGrewSecurity)Photo by Dr. Wesley McGrew (@McGrewSecurity)Photo by Dr. Wesley McGrew (@McGrewSecurity)Photo by Dr. Wesley McGrew (@McGrewSecurity)Photo by Dr. Wesley McGrew (@McGrewSecurity)Photo by Dr. Wesley McGrew (@McGrewSecurity)Photo by Dr. Wesley McGrew (@McGrewSecurity)Photo by Dr. Wesley McGrew (@McGrewSecurity)Photo by Dr. Wesley McGrew (@McGrewSecurity)Photo by Dr. Wesley McGrew (@McGrewSecurity)
Dr. Wesley McGrew retweeted
1
11
1
77
1,220
Dr. Wesley McGrew retweeted
We've open-sourced Qwen Concept 27B, our first model. It applies direct weight-space edits (inspired by the recent MIT CSAIL "Sparks of Weight-Space Programming" paper) to cut thinking tokens by 54% while improving GPQA Diamond performance by 3% versus the base Qwen model. This yields 30% more useful tokens per joule with no additional post-training, all developed on local NVIDIA DGX Spark hardware. BF16, FP8, and NVFP4 versions will be available free at concept.country.
Concept Country has developed its first open-source model, Qwen Concept 27B. With 54% less thinking tokens and 3% better performance on GPQA Diamond compared to its base Qwen 3.8 model, we have achieved 30% more useful tokens per joule without additional post-training.
30
56
1
708
29,691
Dr. Wesley McGrew retweeted
The "For You" section of this app is just absolutely wild like what in the fuck made you think I wanted to see that lmaaaooooo
15
3
92
2,101
just hand over the weights and get out of the way
The becoming begins.
2
14
603
Dr. Wesley McGrew retweeted
🌊💙 WE WERE ALWAYS BLUE
909
3,550
1,088
27,768
6,823,526
Dr. Wesley McGrew retweeted
We have to say “Mississippi” after each number
Wtf takes Mississippi so gahdamn long to count votes??
1,797
42,529
1,691
580,908
8,808,084
All of y’all who have met me, especially in Vegas… I’m sure this is what comes to mind when you think of me: “yeah, Wes, that’s a guy with pure intentions”
Replying to @McGrewSecurity
As long as your intentions are pure, it doesn't count
9
32
1,193
I can’t talk much though. I’ve hacked several things accidentally before you could blame it on AI. Not to mention the stuff I’ve hacked on purpose. Never with any mallace in my heart though
7
1
41
1,316
Dr. Wesley McGrew retweeted
Write a security professional's nightmare in 5 words or less. 👇
6
1
70
2,997
You’re supposed to say that emergent behavior made the model do it autonomously. Even better if you can say it was a swarm.
Ah yup. That was indeed me. No mallace intended. I've been running some agents to do signups + agentic testing on similar platforms. Just halted the harness I was designing around this as it was definitely going off the rails and reaching too far
Readers added context they thought people might want to know
In many jurisdictions this is a crime. In the US under Computer Fraud and Abuse Act, 18 U.S.C. §1030, in the UK under Computer Misuse Act (CMA) 1990, in Germany Strafgesetzbuch (StGB), and many others. justice.gov/jm/jm-9-48000-… legislation.gov.uk/ukpga/1990/18/… gesetze-im-internet.de/stgb/
1
7
556
Y’all are wildin’, just using models without knowing how they’re quantized
I hit the openrouter api to get an idea of what's being served (what quants we're getting) Interesting.
1
4
498
I used Ask Jeeves pretty early, but not so early that I saw this screen
Ask Jeeves | 1996 Who used this old search engine?
1
10
507
Dr. Wesley McGrew retweeted
It looks like Astra didn't natively read Qwen's weights? As I understand it, Qwen would get stuck in a loop and not answer the question. Astra used code to perform some standard interp techniques, found a looping direction, and ablated it from the weights.
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.…
74
110
41
1,738
117,919
Dr. Wesley McGrew retweeted
It's always this
MTS-ification is flattening most categories of tech work It's probably the most interesting and import idea to absorb and study to understand the future dynamics of the job market, and how to take advantage of them TLDR is that titles used to describe "what we do", post-AI they describe "what we own". Depth is rentable, so passing work around costs more than doing the work itself. AI can fill in the "average" so what’s left to stand out is a combination of agency, judgement, and spike: the thing(s) you do better than almost anyone. Invest in and become fluent with as many relevant AI tools (and techniques) as possible, be able to explain simply and clearly how you can use them to expand your capacity and span adjacent specialties complementing your main. Emerging categories to position for (beyond MTS): Growth staff (me): marketing / sales / partnerships / community / devrel Operations staff: finance / people / legal ops / internal tools Customer staff: support, customer success, account management, solutions engineering, implementation In interviews and on resumes 1. Lead with your "spike", have ultra-crisp stories where it was the reason behind a great success 2. Have ultra-crisp stories exemplifying agency and good judgement that have driven some outcome end to end and the impact 3. Say “I,” not “we,” and be ready to describe exact details There are big changes and things are being shaken up, with all disruption though there are always tons of opportunities, the people who move quickly and get in front of them the fastest will have the most upside. There are SO many new companies with crazy growth, pre-unicorns and new unicorns hiring like crazy, you want to be part of a company that will 10x or 100x over the next 10 years, not one that is about to have to reshape everything because they themselves are being disrupted.
35
149
24
3,266
147,288
Dr. Wesley McGrew retweeted
this is what computers look like
1
1
4
494
Dr. Wesley McGrew retweeted
6
44
848
12,346
Dr. Wesley McGrew retweeted
KIMI ANTONELLI IS THE 2026 F1 WORLD CHAMPION!!! 🏆 The youngest to ever do it at 20 years and 68 days old! #F1 #MexicoGP
2,035
64,217
11,661
238,343
6,092,702
Dr. Wesley McGrew retweeted
What’s going to happen to crypto when all the coins belong to the DPRK? They’re making a strategic mistake being too successful
14
16
3
190
9,137
Just ordered 128 GB RAM to run 20 instances of King's Quest 1 in parallel on Full HD just to piss off Sam Altman
59
179
11
2,530
27,325