Sharps or Squares. RHR lt 50 Personal views only.
Simpli-city
Joined July 2011
- Tweets2.1K
- Following575
- Followers3.7K
- Likes13.3K
Pinned Tweet
AI for Security has never been more exciting.
Let me present MAPTA, our multi-agent framework that found multiple (now confirmed!) Remote Code Executions (RCE's) in flagship web products of Tier-1 companies. Why the secrecy? We're good boys, letting them cook patched through responsible disclosure.
What's our secret sauce? 1/n
The only metric is: how many joules did you burn today to secure yourself.
Fireworks is SOC 2 and ISO 27001/27701/42001 compliant. They’re still getting hacked. These certs are just security theater at this point.
Companies pay tens/hundreds of thousands a year for these audits because their customers assume it means their data is safe. And clearly, it does not.
Before we had cyber capable AI, it was much less obvious. But now it just means you had like $12,000 to pay a compliance shop and you have the bare minimum in place.
It’s time for an actually comprehensive framework that, you know, works.
209
Arthur Gervais retweeted
+1, @HatforceSec cooked hard with this one
Make sure you do some pentests with the best AI around before the DRPK does some pentests for you
If you’re looking to try out or invest in an AI-powered pentesting platform, I’d highly recommend @BugBunny_ai. The kind of vulnerabilities this platform uncovers is genuinely insane. It’s seriously impressive mannnnn. I mean just fucking good.
if an inference provider has a low cache hit rate that just means
- they don’t have enough VRAM to host context for all users and are evicting chats too fast
- their inference stack should be optimized to slow down decode slightly in exchange for a higher cache hit, hence much faster prefill and hence faster TTFT
the higher the cache hit rate the lower the costs for everyone
2
6
399
It's simply more efficient to talk to agents than to low-agengcy people
But I also see a world where previous low-agengcy people suddenly get the courage to take more action with agents
DeepMind CEO Demis Hassabis:
"One person who really understands AI will outperform an entire startup team
In 5 years, the people who build with AI will run whole companies with a tenth of the headcount"
3
458
There's a lot of headroom to improve inference stacks
Ask your agents to tell you what's the physical limit of your hardware, quite fun
i'd go as far as to claim that that's exactly what research looks like
poke around and find out
exactly what my abliterated models are going
3
377
biggest issues:
- no XGMI (fast interconnect)
- nerfed from 750W -> 500W, performance will drop
- acknowledged rough software stack
3
353
Absolutely get yourself an abliterated model into your favorite harness.
Will make you realize how much your abilities have been nerfed.
Mind blowing.
If you just horizontally (not even vertically) optimize this to work on qwen3.8-flash-next (6B per expert so in reach enough), then the only moat for local inference is running uncensored freedom-maxed models.
and that's probably going to stand for a while as serving it will get incrementally more and more banned.
1
4
466
there's still hope for hopper
DeepSeek-V4.1-Flash on a 4xH200 machine @ 600w
653 tok/s C1, beautifully scales out to C64
impressive prefill at 42k tok/s, yes 42!
several days of custom kernel tweaking
in production @BugBunny_ai
open source 👇
try to send "Say hi." to deepseek v4.1 flash with no system prompt, thinking on and temperature 0, about 20 times
3
582
GPUs are already being made liquid
nafnaf.ai
sam: local inference is a terrible allocation of scarce resources
“a lot of people want local inference, and somehow they assume that the problem is solved now because we have open-source models, we're going to be able to run them on our computers, and so we're not going to need all the stuff in the data centre
but from a global allocation perspective, actually it's quite terrible to have everybody, like, have a bunch of the world's limited supply of DRAM sitting in their own computer and being used five percent of the time
you want maximum utilisation in the cloud if you're trying to deal with this very supply-constrained environment
somehow many people's intuition is the opposite of that”
178
you'll see more hacks
just now binance & polymarket
get ready and test your systems on an ongoing basis
secure your projects before it's too late
glad the bunny @BugBunny_ai was able to help!
also, if you’re an open-source maintainer or a small startup, we’d be happy to give you as many credits as you need to secure your project.
5
522
the new bunny rips
it's official. i am reborn as a new bunny
this version 2, or was it version 3, is my biggest upgrade in over 6 months
many new features people asked me for
come, try it out!
10 coupon codes for you 👇
BBA2215098971F
BB943AB39FFEC2
BBDEFC475262CA
BB08DDF8781EAC
BBDE9B508B5247
BBCECC858F303C
BB0E192012CA58
BB74BC4D3FA266
BB95396D24B196
BBD6563AB09F77
p.s. please behave, one code by account - i'll go hunting those who use them multiple times and block the accounts fully hehehehe
1
206