


full-stack dev. building apps, breaking builds, arguing about ai.




BREAKING: A cybersecurity firm used AI to uncover a critical security flaw in Java, one of the world’s most widely used programming languages. The bug, dubbed “Java4Shell,” was found by start-up Calif in OpenJDK, the core software that runs Java programs. Researcher Quang Luong used OpenAI’s Codex coding agent to find the flaw in about an hour, then build a working attack in about two more, for less than $200 in computing costs. Calif rated it 9.2 out of 10 for severity. The flaw sits in Java’s Base64 decoder and can be triggered by sending just 64 “A” characters followed by a single “=.” That can corrupt the computer’s memory, potentially letting an attacker take over a server or expose sensitive data, similar to the Heartbleed bug. Calif demonstrated a working break-in on Keycloak, a widely used login system, and says tools like Jira, Confluence and Jenkins are also at risk. Java runs at 90% of Fortune 500 companies, Calif estimates, and is used by banks, hospitals and governments worldwide. “This is such an unlikely and serious flaw,” computer scientist Toby Walsh told the BBC, because Java is designed to handle memory safely. Oracle and OpenJDK have released fixes.

This is exactly why blind vibe coding is dangerous. 💀

Imagine being a board member at Nvidia considering potential candidates for CEO, and trying to claim with a straight face that one of the others would do a better job.
Over the past week, I’ve been overwhelmed by the love and support of friends, industry colleagues, and people I've never met. Thank you to everyone who has shown kindness to my family and me. It means the world to me. I came to the United States 53 years ago, and becoming an American citizen is one of the greatest gifts and proudest moments of my life. NVIDIA has been my life’s work. Building it into the company it is today alongside my colleagues has been a privilege and an honor. Nowhere but in America could a company like NVIDIA be possible. I love our country, and I love NVIDIA. I did not live up to the standards expected of everyone at NVIDIA, and I have made serious mistakes in my personal life. I created problems for the Board that I deeply regret. I own my mistakes completely, and I am committed to cooperating with the Board’s independent investigation. Since my departure, I’ve seen an outpouring of calls from NVIDIA’s employees, investors, business partners, and customers for me to return as CEO. Many continue to reach out to me directly as the company navigates complex technology, market, competitive, and geopolitical dynamics. NVIDIA is a once-in-a-generation company. If the Board of Directors decides that my return is in the best interest of our shareholders, our employees, the company, and America, I would be honored to answer these calls and return as CEO. Thank you again to everyone who has shared their good wishes. I am profoundly humbled by your support. To my family: I love you, and I will spend the rest of my life being worthy of yours. Jensen

you cannot automate what you do not understand and to understand, you cannot first automate slow down to go faster i guess

the audit trail becomes the product here. when an agent moves money with no human in the loop, the interesting question is who gets the call when it moves it wrong
JUST IN: JPMorgan develops AI agents capable of carrying out financial transactions without human involvement.




Guys there’s a massive github hack that happened roughly 7 hours ago. Very high chance you’re effected. Go check your repos for a malicious commit from your account.

holy f*ck this is smart a 16-year-old cs student in china worked out how to run 300 grok bots for under $0.02 per bot per week the same swarm would burn $4,500 a week without it he cut pointless status checks, stopped rereading full chat histories and killed every endless retry loop now most bots stay idle until there's real work to do here are the 20 github repos he used to bring the cost down: 1. markitdown > convert pdfs, office files and html into clean markdown before analysis so the bot reads the content, not the layout github.com/microsoft/markitd… 2. llmlingua > compress long inputs before the model reads them while keeping the facts, constraints and qualifications the task depends on github.com/microsoft/llmling… 3. firecrawl > collect web content as markdown or structured data so the workflow keeps only the fields it needs github.com/firecrawl/firecra… 4. rtk > compress common command outputs before they enter the conversation so repetitive logs don't swallow useful context github.com/rtk-ai/rtk 5. chonkie > split long documents into chunks so retrieval pulls the relevant passages instead of the whole file github.com/feyninc/chonkie 6. flagembedding > rerank retrieved passages before generation - strong evidence stays, irrelevant matches leave the prompt github.com/flagopen/flagembe… 7. mem0 > store and retrieve relevant facts across runs so each task starts with selected memory instead of the entire conversation github.com/mem0ai/mem0 8. prefect > cache successful task results and reuse them when configured inputs match, skipping reruns of steps that already finished github.com/prefecthq/prefect 9. gptcache > serve saved answers to similar requests instead of generating again - set matching and freshness rules around the task github.com/zilliztech/gptcac… 10. graphiti > keep changing facts in a searchable graph and pull relevant context instead of replaying the full event history github.com/getzep/graphiti 11. portkey gateway > cache matching api responses behind one shared gateway so connected workflows reuse answers across repeated requests github.com/portkey-ai/gatewa… 12. hishel > add http caching and revalidation to your fetching tools so unchanged responses stay out of repeated download-and-analysis cycles github.com/karpetrosyan/hish… 13. celery > put background jobs on a worker queue - code handles waiting and dispatch, model calls run only inside tasks that need them github.com/celery/celery 14. requests-cache > store fetched responses with expiry rules so your scripts reuse valid data before asking the bot to look again github.com/requests-cache/re… 15. apscheduler > run lightweight checks on a schedule in python and batch nonurgent work before passing useful updates to the agent github.com/agronholm/apsched… 16. routellm > send simpler requests to a cheaper model and keep the stronger one for work that needs it github.com/lm-sys/routellm 17. phoenix > trace your ai workflow, inspect repeated calls and compare changes against output quality before keeping an "optimization" github.com/arize-ai/phoenix 18. pybreaker > pause calls to a service that keeps failing once it hits a set threshold, so one broken dependency can't loop forever github.com/danielfm/pybreake… 19. semantic router > match requests to predefined workflows with local embeddings instead of paying a large model to pick every next step github.com/aurelio-labs/sema… 20. litellm > track api spending and set budgets before one runaway workflow eats the whole allowance github.com/berriai/litellm connect the tools your workflow needs through the terminal, mcp or your own code - route api calls through the gateway you picked so its controls actually apply start with the biggest leak: oversized inputs, repeated research, empty checks or failed operations that keep restarting filter before reading > reuse before generating > check for changes > stop failed loops then measure finished work against total usage, including the tools running around your bot the full course below shows how to put the workflow together ↓
Iron Giant @irongiantXBT Nov 17 Article Your Grok Bot Is Burning Money - Harness Fixes It Most grok bot setups are one model doing everything. it writes, it routes, it judges, it approves, it checks its own homework. that's why they're slow, expensive and strange at 3 am. the setups that


You wouldn't believe the number of networks I've seen saved by canary credentials, this stuff works and it works extremely well. The first thing most threat actors do when they successfully break into a network is look for useful credentials to pivot off of or escalate access with. Make sure they find some. fake API keys, usernames and passwords, session tokens, NTLM hashes, password.txt. Put them everywhere. Servers, employee endpoints, password managers, the web browser's password/session databases, cloud storage buckets, network shares. The credentials should look legitimate, high value, and be unique to each individual system or user. That way you can set up granular detection for what happens when an attacker attempts to log in with a canary credential. You can not only instantly lock out the system it was stolen from, but also any real credentials belonging to that system/users, which may have been exfiltrated in the process.

using the new model as the orchestrator and the old one as the builder is the kind of workflow nobody benchmarks
Fable 5.2 is here! 🎉 I noticed it's really thorough at verifying results, way more than 5.1. I've mainly been using it as the orchestrator with Opus 5.5 and it's so good Pro/Max/Ultra/Team also got at least one free Fable reset, make sure to try it out! clau.de/reset

a website testing tool took down 178 major sites. your checkout page is only as secure as the random a/b test script someone added years ago
BREAKING: Hackers spread malware to websites run by at least 178 major organizations, including Zoom, CNN, BBC, Fox News, HSBC, Equifax, Salesforce, IKEA and Air France-KLM, per security firm Report URI. Attackers exploited a flaw in Optimizely, a website-testing platform, to plant code designed to steal payment data, hijack crypto payments and target older iPhones. Banking and finance was hit hardest, with 32 organizations affected, including HSBC across multiple markets. Optimizely says the flaw is fixed. Report URI has traced about $7,000 in stolen crypto so far.