Josh Collyer
statsandbytes.bsky.social
Josh Collyer
@statsandbytes.bsky.social
Principal Researcher @ The Alan Turing Institute 🔬Offensive Security ⚔️ and ML/AI Researcher 👾 Part-Time Academic 🧑🏼‍🎓 Rust Enthusiast 🦀 Author of Stats and Bytes Newsletter - https://statsandbytes.fyi/
Pinned
The first issue of Top 5 Security and AI Reads at statsandbytes.fyi is cut and scheduled. Excited to see how it's received! 🫣 What's the worst that can happen!?
Stats and Bytes | Josh Collyer | Substack
Keeping you up to date with where AI and Security collide. Click to read Stats and Bytes, by Josh Collyer, a Substack publication. Launched 2 days ago.
statsandbytes.fyi
I find it fascinating what folks get up to whilst on a train. I'm sat next to someone who is using ChatGPT to write an entire course for new Teaching Assistants...with this name and company in plane view across every slide!
January 24, 2025 at 4:33 PM
I've just realised SubStack can read newsletters to you in the app so it turns into a podcast of sorts. Interested in folks mileage but I've found it hella' good! 😃
January 19, 2025 at 3:42 PM
The 5 reads for this week are locked in.🔒A tad heavy on the LLM related bits (surprise!) but still shaping up to be a good one! This week's reads will cover taint analysis, cyber security task performance evals, bad generation auditing and fault injection mitigation!
www.statsandbytes.fyi
www.statsandbytes.fyi
January 17, 2025 at 10:10 PM
🎩 Top 5 Security and AI reads is out! This week's installment covers:

HoneyPotNet🍯 - Honeypot neural network layers

varMax ‼️- zero day detection NIDS

FrameFlip 🪙 - Rowhammer Deep Learning edition

LLMCompiler 📃 - LLM compiler optimisations

AI Vs Human Spear Phishing 🎣 - Not all it lost
🎩 Top 5 Security and AI Reads - Week #2
Honeypot Neural Network Layers, Zero-day Detection, Rowhammer Deep Learning Edition, LLM Compiler Optimsation and AI phishing being the same (not better!) as human expert performance.
statsandbytes.fyi
January 12, 2025 at 7:17 PM
The progress of AI agents for science is staggering. I'm optimistic and looking forward to seeing the research around using them for security research, particularly around vulnerability identification and then patching.
January 12, 2025 at 12:14 PM
Reposted by Josh Collyer
Found this absolute gem of a texbook on NLP recently, with context from the last few years

web.stanford.edu/~jurafsky/sl...
January 11, 2025 at 3:37 PM
I'm enjoying researching papers for the top 5 reads this week. We've got honeypot methods applied to neural networks, logit analysis to spot zero-days and low level bit flipping to break models. I'm glad folks are piling in to secure AI from all the angles. Excited to see where it goes!
January 10, 2025 at 4:03 PM
Actively looking for interesting stuff to put into this weeks newsletter has resulted in finding too many cool things!

It is so awesome to see the breadth of research going on at the intersection of AI and Security.
January 8, 2025 at 7:23 PM
Absolutely stunned with the amount of engagement the newsletter side hustle is getting. Nearing 70 subscribers now!
January 7, 2025 at 10:02 PM
The first issue of Top 5 Security and AI Reads at statsandbytes.fyi is cut and scheduled. Excited to see how it's received! 🫣 What's the worst that can happen!?
Stats and Bytes | Josh Collyer | Substack
Keeping you up to date with where AI and Security collide. Click to read Stats and Bytes, by Josh Collyer, a Substack publication. Launched 2 days ago.
statsandbytes.fyi
January 4, 2025 at 4:19 PM
I'm excited to be writing the first installment of the Stats and Bytes newsletter. Short, sweet and hopefully popular! 🤘🏼🫣
January 3, 2025 at 4:13 PM