@Antho_Martineau Ax y and I are going sxx zz .5. They Will re watch way.Www is the ratedand c wtwf acts like wf 2ndw fis r safew and w s the same thing w t the same
Ah dang. Looks like a new telnetd vulnerability on the loose 🪲
Once your done patching up this servers be sure to head downstairs and unplug all the fax machines!
https://t.co/nxEwy9DQ0H
There are so many “hills to die on” and little camps within IT.
Hate to say it, but if you use “rm -rf” instead of “rm -fr”, well, I think you are a psychopath!!!
💻 A while back, I had some concerns about AI. Mostly, my lack of knowledge around a tool that I am using daily. I said it would most likely become my next rabbit hole.
🐰 Well, I have started digging and would love for you to join me!
I’m sharing what I’m learning as I go in two places:
📖 Learn AI With Me (the written lessons + code)
📺 Mike Knows Nothing About AI (the matching video walkthroughs)
Lesson 1 is now live: Foundations
It covers LLM basics, like, really basic, like, "why is the answer different every single time" basic. I try to explain exactly how an LLM differs from software that we are traditionally used to!
If you are curious and want to learn alongside me:
📌 GitHub lesson: https://t.co/4j6mjPrDzt
🎥 YouTube video: https://t.co/cWFcqca3R9
If you watch or read it, I’d love feedback. What felt confusing, what should I simplify, what am I missing?
Lesson 2 which deals with tokens will be along shortly!
Seen a few posts now complaining that OpenAI copied Google with translate. Seems like a pretty mundane topic to copy. If that’s the case I mean, did they both copy Marriam-Websters?
It was a pleasure catching up with you on @MadMoneyOnCNBC, @JimCramer! I loved speaking with you on how @RubrikInc helps our customers achieve cyber resilience and deploy trusted AI agents with confidence.
Two missing characters, one horrible result 😬
Wiz shared a supply chain risk tied to AWS CodeBuild + GitHub that (in the worst case) could’ve ended with an attacker gaining control of your repos.
Here’s what went wrong... 🧵👇
Some CodeBuild projects used a webhook filter based on GitHub ACTOR_ID (GitHub account ID) and compared it using a regex allow-list.
But the regex wasn’t anchored (basically missing ^ and $), so it behaved like a substring match instead of an exact match. 🔓
So in essence, an adversary could… 🧨
🔍 Learn which IDs are “trusted” from the exposed/observable pipeline configuration tied to the project.
👥 Create lots of GitHub identities until one ends up with a numeric ID that contains a trusted ID as a substring. Basically, if an approved ID was 1234, a user with an ID of 584912343939 would word as 1234 is in the string. Wiz used automation to speed this up. 🎰
🧾 Open a PR from that identity.
⚙️ In turn, that triggered a CodeBuild run (because the substring match passes the allow-list check).
🕵️ Abuse the build context to grab GitHub credentials/tokens from the build environment/process.
🔑 Use those creds to wreak havoc on the GitHub Org, setting up a classic supply-chain nightmare 💣
AWS patched this up quickly and found no evidence of exploitation ✅
The lesson 🧯 Treat CI like production:
🔐 least-privilege tokens
🧑⚖️ approval gates for privileged builds
and of course,
💾 Backup + Recovery for your GitHub org (and actually test restores)
If you’re thinking about this from a resilience angle, this is why solutions like Rubrik exist. 🧊🛟
Stay safe out there ✌️
https://t.co/JIrvLUsAYa
https://t.co/kIXdT2raD5
@discoposse@jasonfried No kidding. Didn’t matter how many giant pots and cast iron pans you through on top of that thing in the cupboard. Always came out like brand new!
IOPS are dead.
In AI, tokens are the unit of measurement that actually matters.
A token is a chunk of text. Sometimes a word, sometimes part of one. Models do not read sentences. They read tokens. They generate tokens. Everything is counted in tokens.
That is why performance is tokens per second.
No matter what OpenAI and I talk about there is always one thing that is 100% clear - I am the smartest, most creative person the model has ever interacted with. And every idea I have is just so great!
A lot of teams treat Azure DevOps like it’s disposable 🗑️
Worst case we will just recreate pipelines right?
Repos are versioned by got, shouldn’t be an issue
Boards aren’t that important, we’re good!
🤣Sure. Cool cool cool!
That is until Azure DevOps goes sideways and suddenly:
🤷nobody remembers why a change was approved
❌ half the pipelines fail for “reasons”
⛱️ and the one person who knows how it works is on vacation
✅ Turns out DevOps data is… data.
🔥 And losing it hurts more than you think.
Fun fact: I still misspell “focuses.”
In college, I had a summer job with the Canadian federal government, teaching employees Microsoft Word - right when autocorrect was brand new.
My demo? Adding an autocorrect for focusses → focuses.
20+ years later… my brain never updated. Autocorrect still hasn’t either.