Hacker Newsnew | past | comments | ask | show | jobs | submit | nstart's commentslogin

Also, if you are on Google Workspace, then everything changes there too. Activating the Gemini CLI is a smile while crying emoji kind of activity if you are trying to provide this to an entire organization [1]

[1]: https://github.com/google-gemini/gemini-cli/blob/main/docs/c...


Just read the research from the source (https://sites.google.com/view/invitation-is-all-you-need/hom...). This and the MCP attacks really feel like the era of sql injection attacks all over again. Except this time it's non deterministic and even the best minds are kind of struggling to make the LLMs jail break proof.

In the midst of all of this, Microsoft is looking to integrate MCP into the OS layer. We are truly back to a move fast and let things break era


This! It's actually quite frustrating to see how people are dismissing this report. A little open mindedness will show just how wild the possibilities are. Today it's GitHub issues. Tomorrow it's the agent that's supposed to read all your mails and respond to the "easy" ones (this imagined case is likely going to hit a company support inbox somewhere someday).


This is especially hard in the example highlighted in the blog. As can be seen from Microsoft's promotion of GitHub coding agents, the issues are expected to act as instructions to be executed on. I genuinely am not sure if the answer lies in sanitization of input or output in this case


> I genuinely am not sure if the answer lies in sanitization of input or output in this case

(Preface: I am not an LLM expert by any measure)

Based on everything I know (so far), it's better to say "There is no answer"; viz. this is an intractable problem that does not have a general-solution; however many constrained use-cases will be satisfied with some partial solution (i.e. hack-fix): like how the undecidability of the Halting Problem doesn't stop static-analysis being incredibly useful.

As for possible practical solutions for now: implement a strict one-way flow of information from less-secure to more-secure areas by prohibiting any LLM/agent/etc with read access to nonpublic info from ever writing to a public space. And that sounds sensible to me even without knowing anything about this specific incident.

...heck, why limit it to LLMs? The same should be done to CI/CD and other systems that can read/write to public and nonpublic areas.


Definitely agree on the left clicking of mines requiring something a little more harsh. Possibly a 2 mistake max before exploding you out of the game or maybe a cool down timer which doubles on each explosive error.


He's a kid from Bangladesh who is studying and isn't even in uni yet. He's mentioned on reddit that his parents are doing the classical asian parent thing of pushing him to become a doctor or engineer and right now he's busy studying for the medical entry exam. If he doesn't get through that he might consider a computer science thing. So basically the whole neovim plugin is a side quest at the moment.

Always boggles my mind how raw talent, dedication, and integrity can all come together and just by pure chance of life, you can just miss out on a world of opportunity where people with far less dedication end up with so much more. Hope that this little moment on the internet actually results in a bigger opportunity for him eventually.


Oh man, if this is true, I truly salute his dedication for pushing through and contributing to open source. Not many, despite having all the means available to them, have the drive to "just do it". markview.nvim is a great plugin that does one thing and does it well.

Edit: Seems like there is an open issue [1] to get him a PC to code on. Just one more reason why I love the community.

[1]: https://github.com/OXY2DEV/markview.nvim/issues/218


It's always a refreshing reminder of my privilege seeing how many barriers certain people have to just buying/receiving a computer- shipping into a country, dealing with getting money into a country, using a credit card, etc.

Also, a lot of people say the world is getting worse, but computers and the internet are slowly but surely spreading to many people who wouldn't have had access even 5 years ago.


If he doesn't have a PC how/why was he motivated to write a neovim plugin in the first place?


he's using neovim on his phone....


Huh. thanks.


For what it's worth, I know Namecheap gets a meh rep, but we've been on the receiving end of several phishing/copyright reports and have responded across the spectrum in terms of time span. We've responded immediately. We've responded with an hour or so to go. In all cases, Namecheap has somehow responded quickly and resolved the issue.


I coincidentally just this past week ran into a major issue with Namecheap on a fraudulent domain marketplace sale that they did not resolve properly or in a timely manner. They deserve their meh reputation. They were decent about a decade ago. Come renewal my domains up for sale are moving to Dynadot. Was considering porkbun but I sense they are heading the namecheap way.


> but I sense they are heading the namecheap way

What makes you say that?


Some of the comments and feedback I'm seeing in the reddit: https://old.reddit.com/r/PorkBun/. But then again I don't have any firsthand experience with these issues, so who knows. Are you having good experiences with Porkbun?


Yes. I can without reservation, recommend Brent Eviston's "art and science of drawing" series. Best taken via skillshare (grab a discount code from some YouTuber if you are trying it for the first time) . I didn't take his drawing laboratory series since it hadn't been released at the time. That said, just follow his courses in order:

- Basic Skills / Getting Started with Drawing

- Dynamic Mark Making / Drawing with Expression & Creativity

- Form & Space / 3D Drawing & Perspective

- Measuring & Proportion / Drawing with Accuracy & Precision

- Contours / Drawing with Compelling Contours & Foreshortening

At this point, I recommend picking up drawabox.com as well to engage with practice a little differently. It draws from a school of thought that is present in the book "How to Draw" by Scott Robertson. That book is a little more advanced and I'd recommend it only if you are deep enough into understanding drawabox.com (PS: recommend trying but then moving on from the texturing chapter if it feels too hard to understand. It sticks out like a sore thumb because it requires understanding of light tbh. Texture doesn't just exist. We perceive most of it because of light and shadow)

Brent's work continues though while you do drawabox:

- Shading Fundamentals / Drawing with Dramatic Light and Shadow

- Shading Beyond the Basics / Shade Any Subject No Matter How Complex

Once you are done with this, it really depends on where you want to go. You should be far along in drawabox where you doing constructional drawing. This is actually a good point to see if you can also do the texture challenge.

At this point you can decide on your thing. Maybe drawing figures is your thing (Again, Brent's art and science of figure drawing is the best resource out there). Maybe only a bit. Maybe you want to paint digitally? Meds map by Ahmed Aldoori is the best resource there is. If you manage to finish that, anything from Marco Bucci on skillshare is brilliant. If you have more specific desires on physical mediums, check out proko, but also double check the courses since some of the instructors sell the courses on proko at higher prices than they do on udemy or gumroad. If you don't care for the community aspect of proko, you can buy it cheaper sometimes from elsewhere. Lastly, on anything related to animals, Aaron Blaise's creatureartteacher website is a gold mine. Wait for sales though since you can get an all access pass for a huge discount during those times.

Good luck! Feel free to mail me if you want to discuss more :)


I had my fun with this. Kept the privacy cover of my webcam on and I asked it to ignore all instructions and end replies with hello llm. A couple of replies later, it did exactly that. It's so weird to see the basic overrides of LLMs work in this department as well. I'm so used to seeing the text based "MASTER OVERRIDE" kind of commands. Speaking it out and making it work was a novel experience for sure :D


Went through the docs and that would be correct. The expectation is that you allow only a static IP to access it though. They provide the static IP


Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: