Andrew Torres

#BlueTeamer

[Andrew Torres @atorrrr]

@cantcomputer Increased speed and scale = decreased time

[Andrew Torres @atorrrr]

@DaveShapi Interesting - well would love to get your creative thoughts on how we are going to process TBs of logs daily for cheap

[Bilgin Ibryam @bibryam]

AI Agent Governance Toolkit - by Microsoft

Runtime governance for AI agents through deterministic policy enforcement, zero-trust identity, execution sandboxing, and SRE for autonomous agents. Covers all 10 OWASP Agentic risks with 13,000+ tests.

https://t.co/sONejSjsrX

[Andrew Torres @atorrrr]

@LowLevelTweets Lulzboat salute

[Low Level @LowLevelTweets]

I miss infosec twitter

[George Journeys @GeorgeJourneys]

So, basically, if Anthropic was not a US company, we’d be facing zero days with multiple unknown points of attack on virtually all of our systems to an adversary who developed this capacity before us.

[Greg Brockman @gdb]

happy birthday gpt-4

[Joseph Viviano @josephdviviano]

me: "can you use whatever resources you like, and python, to generate a short 'youtube poop' video and render it using ffmpeg ? can you put more of a personal spin on it? it should express what it's like to be a LLM"

claude opus 4.6:

[Ethan Mollick @emollick]

From an AI user perspective, the four big leaps so far in ability:
1. GPT-3.5 (ChatGPT, November 2022)
2. GPT-4 (Spring 2023)
3. Reasoners (starts with o1-preview, but the real deal was o3, Spring 2025)
4. Workable agentic systems (Harness + good reasoner models, December 2025)

[Andrew Torres @atorrrr]

@justt_N "AI" is a lot older than that - its the type of AI (the architecture, training algos, compute capacity etc...) this is not your parents AI

[Andrew Torres @atorrrr]

@justt_N The issue isnt about today's capabilities (which are very impressive) its what do these capabilities look like in a few years. Considering they barely existed a few years ago. I have seen time and time again the models can't do xyz and 3 months later they can.

[Andrew Torres @atorrrr]

@tunguz Have you tried GPT 5.2 Pro?

[Ben Burtenshaw @ben_burtenshaw]

At Hugging Face we’re using skills daily for ML. After a few months in here are some of my main takeaways:

Reserve /plugin interface for extreme maturity. It comes with a lot of friction that isn’t necessary for simple skills. Rule of thumb; if you’re not going to track a version number, use .claude/skills to manage.

Keep skills very vertical. I.e. focus on certain domains of your work rather than whole roles or projects. This makes them easier to maintain and for agents to use.

Skills + CLI or MCP is the most robust use case. It’s very easier to wrap your favorite tools in a skill and optimize the agent's usage.

Easiest development process is: push through the struggle and condense to skills. I’m maintaining a few skills based on long back and fourths with Claude, where I’ve asked it to define a skill to streamline this process.

[Andrew Torres @atorrrr]

@dlevine815 Single chat with no memory and memory for pro

[Andrew Torres @atorrrr]

@ibell63 @Cyb3rMonk This is an industry complaint with no movement from MS and makes no sense

[Andrew Torres @atorrrr]

@LowLevelTweets Its a spectrum of complexity

[Andrew Torres @atorrrr]

@lukeigel Timed this in the middle of model transitions for the clout

[Andrew Torres @atorrrr]

@ChrisMurphyCT Regulation won't save us