November 21, 2023
November 21, 2023
This never gets old. pic.twitter.com/zwzsA3JjVs
— Doctrine Man (@Doctrine_Man) November 20, 2023
tmp.0ut 3 is out!
Check it out, it's tmp.0ut Volume 3!https://t.co/Gm9erN8hTJ pic.twitter.com/gUTaS4SYjt
— tmp.0ut (@tmpout) November 20, 2023
This sea turtle snuggles into a sea sponge and lets out a big yawn before a nap 🐢 pic.twitter.com/MZzP9OWk1o
— Nature is Amazing ☘️ (@AMAZlNGNATURE) November 21, 2023
Can’t help but feel that data analysis would really improve the impact of ransomware leaks. Basically, they would search for actual narratives to expose and prepare packages of those with supporting evidence to make public. Just like any leak operation, you need to make a press packet to increase the chances of coverage, particularly coverage that you want to see.
Create narratives with the data and make sure they are newsworthy. So, scandalous or outrageous or illegal etc. etc.
Dumping huge amounts of data just means that no one will comb through it for stories.
https://www.theregister.com/2023/11/10/lockbit_leaks_boeing_files/LLMs cannot find reasoning errors, but can correct them!
[2311.08516] LLMs cannot find reasoning errors, but can correct them!
While self-correction has shown promise in improving LLM outputs in terms of style and quality (e.g. Chen et al., 2023; Madaan et al., 2023), recent attempts to self-correct logical or reasoning errors often cause correct answers to become incorrect, resulting in worse performances overall (Huang et al., 2023). In this paper, we break down the self-correction process into two core components: mistake finding and output correction. For mistake finding, we release BIG-Bench Mistake, a dataset of logical mistakes in Chain-of-Thought reasoning traces. We provide benchmark numbers for several state-of-the-art LLMs, and demonstrate that LLMs generally struggle with finding logical mistakes. For output correction, we propose a backtracking method which provides large improvements when given information on mistake location. We construe backtracking as a lightweight alternative to reinforcement learning methods, and show that it remains effective with a reward model at 60-70% accuracy.
TSA facial recognition bad, etc etc
Don’t Take It at Face Value: Why TSA’s Implementation of Facial Recognition is More Dangerous Than You Think – EPIC – Electronic Privacy Information Center
Facial recognition is an invasive and dangerous surveillance technology. When the government moves forward with pilot programs that will, if fully implemented, subject millions of people on a daily basis to the technology that should give us all pause.
Following #POC2023 I did some analysis on CVE-2023-36911 (x64 & x86). I looked at more recent and older versions (small differences). I'll have a short write-up later but here is a basic repro based on what was disclosed by KunLun Lab in the talk ❤️ https://t.co/qsecINolwi
— b33f | 🇺🇦✊ (@FuzzySec) November 21, 2023
tbh seems like there are easier ways to get a job at microsoft
— Nick Carr (@ItsReallyNick) November 20, 2023
Skyview
I’m training the new guy at work how to gradually lose the will to live
Skyview
I'm dead. 💀💀💀
“When I said you were to go submarine hunting, I didn’t mean in first-person” pic.twitter.com/sFeTXLIQ3y
— H I Sutton (@CovertShores) November 21, 2023
Rare to capture images of them spawning
— Dreadnought Holiday (@TheDreadShips) November 21, 2023
"Dive"
— Alessandro Ponzetto (@AlessandroPonz4) November 21, 2023
"But sir we're a plane"
"I said dive"
this japanese restaurant has a vegan version of their noodles & instead of taking a new photo they just censored out the egg pic.twitter.com/ssW2mFHezT
— gaz (@gazpachomachine) November 20, 2023