• 4 Posts
  • 1.2K Comments
Joined 3 years ago
Aquileo | cake
Cake day: July 4th, 2023

Aquileo | help-circle




  • It should theoretically be impossible with the way you set it up.

    You explicitly, opt in, expose which parameters are available to the LLM, it cant invoke any params beyond that.

    And you can set params to hard values that arent exposed.

    Finally, you can use the enum config to further specify a whitelist of valid values for a param, anything else gets rejected.

    This pretty much locks most tools down tight.

    Though, wouldn’t hurt to add integration tests covering scenarios like this explicitly.



  • The tl;dr:

    1. Prompt inject a malicious instruction in a word document that instructs the AI to copy this instruction to other documents as part of the payload.

    2. Dumb user downloads and opens the document with copilot enabled, abd ignores the large suspicious white blank page that totally doesnt look like a hidden giant injection attack.

    3. Thats it pretty much it.

    Copilot will get injection attacked because the prompt is super huge and at the end of the document, so its prior instructions start to fuzzy out.

    Then it’ll go “okey doke” and start copying the prompt injection attack payload to a bunch of other documents.

    The fix is stupid simple… copilot should just be prompting the user for permission if it ever edits a file other than the one that is open. Im surprised that isnt already the case…?

    It certainly is already the case for copilot in vscode.


  • This is pretty much entirely why I built https://github.com/SteffenBlake/cli-mcp-mapper

    I map bespoke CLI commands, with parameters, to MCP tools, and then 100% just turn off the ability to invoke the CLI itself entirely.

    That way it can only explicitly invoke the whitelisted commands I have granted it access to run, with only the parameters I have allowed it to.

    So for example I can explicitly expose git commit, status, stage, unstage, but everything else git related is not available, not even git push or pull.

    This substantially seems to help prevent the LLM from going off the rails on stupid CLI nonsense, like when it starts cating files instead of just fucking reading them.




  • The real reason though is simple:

    When you are younger, your center of balance is way more core and your arms/legs weigh a lot less.

    As you grow your limbs become way heavier.

    Think about how easy it is to pick up a 7 year old vs a 30 year old.

    Well, when you do a cartwheel, you are also having to move the body of a grown adult, your own body that is.

    So its possible if you are very fit, of course.

    But for an average person, good luck haha.

    General rule of thumb is that a cartwheel is similiar to a pull up.

    Remember how easy pull-ups were when you were a kid? Same principle.


  • These terms were originally heavily used as heavy right wing 4chan terms, largely circulating in hate circle jerks (which is like 50% of 4chan anyways). Largely associated with millenial incel culture.

    Gen Z proceeded to take these terms and parody the fuck out if them, effectively mocking incel culture and turning it into a huge joke. Thus then became mainstream and its largely acted as a very successful humiliation tactic towards the hateful groups.

    Think stuff like all the “chad” memes but applying it to silly stuff, like intense cuts of a random chair or a porta-potty, x-pilled terms, etc.

    All their terms have been mocked into oblivion, you cant really use the terms without just sounding like a gen z meme now.

    An example of when millennials did a similiar thing was mocking the shit out of gen x biker/trucker hyper masculine culture and turning it into an ironic joke. IE the skeleton “my body is a machine that turns x into y” jokes, or ironic memes of edgy skeletons/wolves/etc. Three moons shirt, those things.

    10/10, I fucking love how culture can do stuff like this over time lol




  • pixxelkick@lemmy.worldtoAI CodingMy Agentic Coding Setup, July 2026
    Aquileo | link
    Aquileo | fedilink
    English
    Aquileo | arrow-up
    3
    Aquileo | arrow-down
    2
    ·
    9 days ago

    I have a similiar setup, but streamlined way more.

    1. I just use docker on my Linux dev box to create disposable containers that my agents can do work sandbox’d inside of. But, Id like to eventually try out Amazon’s firecracker instead as a way to create thin linux VMs on the server machine.

    2. I use Github Copilot as my AI distribution subscription, as this has first class integration with VS Code and allows me to easily rollback code changes, and click to preview inside of vscode natively.

    3. I lock down all commands my agents can run behind MCP, instead of “can do anything” access to the terminal, my agents have an explicit set of MCP tools that wrap terminal commands, with params. This dramatically improves keeping the agent on the rails and avoids common issues like it getting very grep hungry and prepping files constantly (I return a 1 line warning anytime the agent invokes the grep tool on a file less than 200 lines in length, instead of results)

    This simplifies a lot, I can manage a preset set of plugins and etc installed on the vscode instance via a dockerfile + docker conpose, letting me define a bespoke set of installed tools for a given project.




  • pixxelkick@lemmy.worldto"Quotes"@quokk.auYoung Men
    Aquileo | link
    Aquileo | fedilink
    Aquileo | arrow-up
    1
    ·
    10 days ago

    In my country, a bit more than half are.

    However the point Im making is if I asked a random collection of young men here to identify which is which, most of them would get most of the answers wrong, and theyd argue over it.

    Because theres basically zero education on the topic. Not exactly there fault per-se, Im moreso calling out the government for failing our youth on this.

    Ill also note that most young women would also probably get a lot of the answers wrong too…