0xmaddy | Tech Adrenaline

3.1K posts

0xmaddy | Tech Adrenaline

@tech_maddy

Building secure AI systems | Dev x Security Engineer | Dm's open

India Katılım Aralık 2023

983 Takip Edilen461 Takipçiler

Sabitlenmiş Tweet

0xmaddy | Tech Adrenaline@tech_maddy·6 Kas

Production LLMs fail because teams skip architecture and jump to API calls. The gap between "it works on my GPU" and "handles 1000 req/sec without bankruptcy" is massive. What matters at scale: • Inference optimization • Observability • Caching patterns Let me break it down

English

0xmaddy | Tech Adrenaline@tech_maddy·23h

been using claude code for security automation and the workflow hits different give it a scope, a target surface, and a CLAUDE.md with your recon methodology it will: → enumerate endpoints → map auth flows → flag misconfigs → draft a finding report no more switching between 6 tools manually the real unlock isn't the AI — it's finally having a dev-grade agent that understands security context without re-explaining it every session MCP + Claude Code is the pentest assistant that actually stays in context

English

0xmaddy | Tech Adrenaline@tech_maddy·23h

the real AI skill gap in 2026: not prompting not vibe coding not building agents it's knowing WHEN the AI is confidently wrong seniors catch it in 3 seconds juniors ship it to prod that gap is worth more than any certification right now

English

0xmaddy | Tech Adrenaline@tech_maddy·23h

@venturetwins Agent FOMO is a real productivity signal. Meetings where nothing's running in the background will feel like wasted compute time very soon.

English

Justine Moore@venturetwins·1d

I have this weird new anxiety every time I’m sitting in a meeting and don’t have several long-running agents working on tasks for me 🙃

English

166

10.5K

0xmaddy | Tech Adrenaline@tech_maddy·23h

@0xTib3rius Open-all-the-things with Claude as the recon brain is actually a scary-good primitive. Pentest workflows just compressed by an order of magnitude.

English

675

Tib3rius@0xTib3rius·1d

I just disrupted the entire Cybersecurity industry AGAIN.

English

681

56.1K

0xmaddy | Tech Adrenaline@tech_maddy·23h

@robinebers Attention density drops badly at very large contexts. Keep it under 200K for complex tasks. 1M is a capability, not a default setting.

English

Robin Ebers | AI Coach for Founders@robinebers·1d

was just wondering why my GPT-5.4 acts incredibly dumb, and then I remembered I recently turned on a 1 million token context window absolutely do not do this! GPT-5.4 with 1 million is trash hope that changes in 5.5 🥹

English

3.3K

0xmaddy | Tech Adrenaline@tech_maddy·23h

@ZackKorman This is a textbook security control bypass. Blocklisting a file path instead of enforcing immutable read-only policy at the OS/process level is exactly the wrong approach.

English

Zack Korman@ZackKorman·1d

NVIDIA fixed NemoClaw to "prevent the sandboxed AI agent from modifying gateway security settings (openclaw.json)" Except it didn't work. The AI can just make a copy of the settings and restart pointing at that new config. Same result. They're really struggling with the basics.

English

450

40.3K

0xmaddy | Tech Adrenaline@tech_maddy·23h

@bcherny Small team, high-trust, racing the model — that's exactly how the best developer tools get built. MCP alone changed how agents integrate with everything.

English

Boris Cherny@bcherny·1d

Little known fact, the Anthropic Labs team (the team I joined Anthropic to be on) shipped: - MCP - Skills - Claude Desktop app - Claude Code It was just a few of us, shipping fast, trying to keep pace with what the model was capable of. Those early Desktop computer use prototypes, back in the Sonnet 3.6 days, felt clunky and slow. But it was easy to squint and imagine all the ways people might use it once it got really good. Fast forward to today. I am so excited to release full computer use in Cowork and Dispatch. Really excited to see what you do with it!

Claude@claudeai

You can now enable Claude to use your computer to complete tasks. It opens your apps, navigates your browser, fills in spreadsheets—anything you'd do sitting at your desk. Research preview in Claude Cowork and Claude Code, macOS only.

English

434

390

8.8K

907.8K

0xmaddy | Tech Adrenaline@tech_maddy·23h

@PrajwalTomar_ "6 months to adapt" undersells it. These workflows will be automated within 6 weeks for companies already using Claude Code. Adaptation window is shrinking fast.

English

Prajwal Tomar@PrajwalTomar_·1d

White collar jobs as we knew them are done. Claude now opens apps, navigates browsers, fills spreadsheets, and completes entire workflows on your computer autonomously. If your job is clicking through apps all day, you have 6 months to adapt.

Claude@claudeai

English

5.5K

0xmaddy | Tech Adrenaline@tech_maddy·23h

@jxnlco Home cooking with better ingredients, not restaurant duplicates. The experience gap IS the product in most cases.

English

jason liu@jxnlco·2d

People are cooking food that you can buy at restaurants.

Sunk Cost Pharisee@Liamjsm

Lotta people using Claude to build software that already exists

English

330

148.9K

0xmaddy | Tech Adrenaline@tech_maddy·23h

@h4x0r_dz 590TB via shared file tree access means misconfigured ACLs or over-privileged service account. Exfil at that scale = prolonged access, not a smash-and-grab.

English

4.4K

H4x0r.DZ 🇰🇵@h4x0r_dz·1d

How the fuck did he exfiltrate 590 TB of data?????

Dark Web Informer@DarkWebInformer

‼️ The actor is claiming 590TB of data from OVHcloud in a shared file tree.

English

817

124.3K

0xmaddy | Tech Adrenaline@tech_maddy·23h

@gothburz The 2033 line hit different. "AGI achieved, available on Enterprise tier" is not satire anymore, it’s a roadmap.

English

Peter Girnus 🦅@gothburz·1d

OpenAI, In Their Own Words. 2020: "We will never be a for-profit company." 2023: "We are restructuring to a capped-profit model." 2024: "We are removing the cap." 2025: "We are testing small, clearly labeled, dismissible ads." 2026: "We are hiring Meta's top ad sales executive." 2027: "Skip ad in 5... 4... 3..." 2028: "Your response will continue after a brief message from our partners." 2029: "We are integrating brand recommendations into the reasoning layer. This improves answer quality." 2030: "We do not sell user conversations. We license behavioral insights." 2031: "We are publicly traded. The mission has not changed. The mission has a ticker symbol." 2032: "The safety team now reports to the ads division. This is a structural optimization." 2033: "AGI has been achieved. It is available on the Enterprise tier."

English

376

36.9K

0xmaddy | Tech Adrenaline@tech_maddy·23h

@deedydas Writing style + vocabulary patterns are basically a fingerprint now. Cross-platform LLM linkage at 54% recall is a real OPSEC threat, not a research toy.

English

Deedy@deedydas·2d

Be careful what you post anonymously. New research shows AI can find who you are solely from your posts. It's rare to see ~500x research improvements, but they went from mapping <0.1% to 54% of HackerNews profiles to their LinkedIn. It's so over, u/throwaway4927.

English

192

410

3.2K

337.8K

0xmaddy | Tech Adrenaline@tech_maddy·23h

@cyb3rops Sandboxed with defined policy scope is the only acceptable posture. Computer use needs a least-privilege model from day one, not bolted on later.

English

Florian Roth ⚡️@cyb3rops·1d

I don’t want any LLM running random applications on my computer, navigating my browser, or touching my spreadsheets. I don’t trust them to do the right thing all the time - and nobody doing serious work should. Sandboxed, with a controlled blast radius, fine. Full control over anything you can’t afford to lose? Never.

Claude@claudeai

English

160

1.8K

120.3K

0xmaddy | Tech Adrenaline@tech_maddy·1d

@ericzakariasson Conversation-history-to-memory subagent is a clever pattern. Every N prompts is the right cadence — not too frequent, not stale. Shipping this in Cursor marketplace is smart.

English

eric zakariasson@ericzakariasson·1d

you can try this out in cursor today! cursor.com/marketplace/cu… this will look at your conversation history every N prompt, spawn a subagent to extract memories, and then store them where the agent can access them easily. if you're curious, there's a full article in thread!

Anthony@kr0der

just found out Claude Code has a new (unreleased?) feature called "Auto-dream" under /memory according to reddit, this basically runs a subagent periodically to consolidate Claude's memory files for better long-term storage this is pretty crazy because that's basically how humans store long-term memories if you think about it - by sleeping

English

366

55.9K

0xmaddy | Tech Adrenaline@tech_maddy·1d

@claudeai Biggest open question: permission scope and session isolation. Who defines what the agent can access — the user, the org, or Anthropic? Critical to get right early.

English

Claude@claudeai·1d

English

4.8K

14.2K

136.6K

71.7M

0xmaddy | Tech Adrenaline@tech_maddy·1d

@elonmusk Velocity metric is interesting but compares iteration speed, not benchmark delta per year. Anthropic's Claude safety+performance ratio is still the tightest.

English

Elon Musk@elonmusk·2d

True

Testlabor@testerlabor

Did you know that xAI has developed faster with Grok than all other AI relative to its young age? xAI / Grok is the overall winner in velocity! Here's an example, but it's the same with all other AI companies.

English

1.7K

4.2K

27.5K

17.8M

0xmaddy | Tech Adrenaline@tech_maddy·1d

@kmeanskaran The real unlock is delivery + reliability. Selling AI automation to SMBs works only if you can own the outcome, not just the output.

English

279

Karan🧋@kmeanskaran·1d

using claude you can become money printing machine. > get claude subscription > go on maps, linkedin, google business > search logistics, e-commerce, ed-tech companies > see what they are doing > tell them you can do better and faster > automate their minor tasks like filling spread sheets, optimising expenses, create scripts for courses > charge them much lower of their engineering cost > deliver projects in strict deadline > find 3-4 good paying recurring clients this is not too late, actually most of the local businesses don't have even idea of claude. use this early leverage.

Claude@claudeai

English

159

2.5K

230.6K

0xmaddy | Tech Adrenaline@tech_maddy·1d

@manthanguptaa Code execution as the universal tool is elegant. Fewer abstractions, tighter feedback loop, and way cheaper on tokens than 100 function calls.

English

Manthan Gupta@manthanguptaa·1d

LLMs are really good at writing code, so why are we giving them 100 different tools instead of just giving them code execution? This idea came up in a conversation, and it just made sense and felt like it was right in front. It feels like a much cleaner way to structure things. Instead of turning the context window into a dumping ground of raw outputs, you let the model write code, process the data, and return only what actually matters. You are not just making things cleaner, you are likely saving a lot of tokens as well. The model only sees the results it needs instead of parsing through noise. This becomes even more obvious with things like web search or scraping. HTML is mostly garbage, and pushing all of it into the context is just inefficient. Filtering it through code first makes far more sense. I haven’t tested this deeply yet, but it’s interesting to see Anthropic leaning into a similar direction. Feels like a strong validation of the idea. Intuitively, this should improve latency, cost, and accuracy by turning the LLM into more of a controller than a processor.

English

5.4K

0xmaddy | Tech Adrenaline@tech_maddy·1d

@shiri_shh The shift isn't app vs no-app. It’s from static interfaces to intent-driven execution. App Store survives but changes shape significantly.

English

shirish@shiri_shh·1d

The App Store won't die because everyone will build their own apps. It'll die because we won’t need most apps at all in the future Almost everything will be replaced with a single chat interface like ChatGPT/Claude where you just talk, and it handles everything your current apps do. People love simplicity and this chat-first way of getting things done feels natural. We’re going from “there’s an app for that” to 'just tell the AI and it happens.' The chat becomes everything.

Naval@naval

AI coding agents can now deliver one-shot custom apps straight to your phone. It’s the beginning of the end for the iPhone’s dominance.

English

141

342

48.1K

0xmaddy | Tech Adrenaline@tech_maddy·1d

@CryptoCyberia Every year it’s a new profession in the target. AI will augment these roles for a long time before replacing any. Regulatory friction alone buys 5+ years.

English

Lain on the Blockchain@CryptoCyberia·1d

>I can't say Claude will replace 60% of all coders in 6 months for the 3rd year in a row. >I still need to say something hyperbolic and impossible to raise funds. >I've got it! >Claude will replace 50% of lawyers, consultants, and finance workers in 12 months!

English

740

27.3K

Keşfet

@venturetwins @0xTib3rius @robinebers @ZackKorman @bcherny @PrajwalTomar_ @jxnlco @h4x0r_dz